An Articulate Model of the Water Cycle - Multiple Choices ...
An Articulate Model of the Water Cycle - Multiple Choices ...
An Articulate Model of the Water Cycle - Multiple Choices ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>An</strong> <strong>Articulate</strong> <strong>Model</strong> <strong>of</strong> <strong>the</strong> <strong>Water</strong> <strong>Cycle</strong><br />
Using qualitative reasoning<br />
Jaeque V. Koeman<br />
studentnummer 9619291<br />
Universiteit van Amsterdam<br />
Opleiding Sociaal Wetenschappelijke Informatica<br />
Roetersstraat 15<br />
1018 WB Amsterdam<br />
Augustus 2003<br />
Supervisie: Dr. Bert Bredeweg
Contents<br />
CONTENTS.................................................................................................................................................... 3<br />
1. INTRODUCTION....................................................................................................................................... 5<br />
2. KNOWLEDGE SYSTEMS IN EDUCATION: AN OVERVIEW ............................................................. 7<br />
3. QUALITATIVE REASONING PARADIGM.......................................................................................... 10<br />
3.1 DESCRIBING CHANGE ............................................................................................................................ 11<br />
3.2 COMPOSITIONAL MODELLING................................................................................................................ 12<br />
3.3 QUALITATIVE SIMULATION.................................................................................................................... 14<br />
3.4 ONTOLOGY ........................................................................................................................................... 14<br />
4. GARP: A SPECIFIC SIMULATOR ........................................................................................................ 16<br />
4.1 GARP .................................................................................................................................................. 16<br />
4.2 QUANTITIES.......................................................................................................................................... 16<br />
4.3 EQUALITY............................................................................................................................................. 17<br />
4.4 FUNCTIONAL RELATIONSHIPS................................................................................................................. 18<br />
4.5 MODEL FRAGMENTS.............................................................................................................................. 19<br />
5. MODEL CONSTRUCTION..................................................................................................................... 21<br />
5.1 MODEL DESIGN PROCESS ....................................................................................................................... 21<br />
5.2 CLASSIFICATION ................................................................................................................................... 23<br />
5.3 ORGANISATION OF MODELS AND ONTOLOGY .......................................................................................... 24<br />
5.4 QUALITATIVE MODELLING OF A DOMAIN ................................................................................................ 25<br />
6. THE WATER CYCLE ............................................................................................................................. 28<br />
6.1 DOMAIN DESCRIPTION: INTRODUCTION TO HYDROLOGY AND METEOROLOGY. ......................................... 28<br />
6.2 CONCEPTUAL MODEL ............................................................................................................................ 31<br />
6.3 DECOMPOSITION AND MODELS OF DIFFERENT COMPLEXITY..................................................................... 33<br />
6.3.1 Homer........................................................................................................................................... 34<br />
6.3.2 <strong>Model</strong> 1......................................................................................................................................... 34<br />
6.3.3 <strong>Model</strong> 2......................................................................................................................................... 40<br />
7. SIMULATION OF THE WATER CYCLE ............................................................................................. 56<br />
7.1 SIMULATING VERTICAL DIFFERENCES..................................................................................................... 56<br />
7.2 SIMULATING MODEL 1 .......................................................................................................................... 59<br />
7.3 SIMULATING MODEL 2 .......................................................................................................................... 62<br />
8. CONCLUSION AND DISCUSSION. ....................................................................................................... 68<br />
9. ACKNOWLEDGEMENTS ...................................................................................................................... 70<br />
LITERATURE.............................................................................................................................................. 71<br />
3
Abstract<br />
This study describes several qualitative models capable <strong>of</strong> reasoning about <strong>the</strong> behaviour <strong>of</strong> <strong>the</strong><br />
ecological domain <strong>of</strong> <strong>the</strong> water cycle. Qualitative models are useful in terms <strong>of</strong> explicit knowledge<br />
communication, and <strong>the</strong>y can form a component <strong>of</strong> an intelligent tutoring environment. Particular<br />
methodological aspects <strong>of</strong> <strong>the</strong> design <strong>of</strong> qualitative models support <strong>the</strong> construction <strong>of</strong> simulation<br />
models that are suited for didactic purposes. Given this purpose, conceptual knowledge is captured in<br />
model fragments, which describe structures and processes. The domain <strong>of</strong> <strong>the</strong> water cycle can be<br />
viewed and explained with various scopes. Following ideas on how to structure <strong>the</strong> communication <strong>of</strong><br />
knowledge, qualitative models inhabiting different levels <strong>of</strong> detail and perspective have been<br />
developed, each resulting in a particular simulation. The simulations are discussed and analysed,<br />
focusing on knowledge communication with <strong>the</strong> use <strong>of</strong> computer systems that are based on <strong>the</strong>ories<br />
from cognitive science, artificial intelligence and instructional science.<br />
4
1. Introduction<br />
When interacting with <strong>the</strong> physical environment, people need knowledge <strong>of</strong> <strong>the</strong> behaviour <strong>of</strong> systems<br />
and processes. The specific type <strong>of</strong> knowledge that people need depends on <strong>the</strong> tasks <strong>the</strong>y have to<br />
perform in <strong>the</strong> environment, e.g. controlling, constructing or diagnosing, see also (Breuker & Van de<br />
Velde, 1994), and <strong>the</strong> particular nature <strong>of</strong> <strong>the</strong> domain being dealt with (e.g. <strong>the</strong>rmodynamics or<br />
biology). The <strong>the</strong>sis presented here concerns <strong>the</strong> development <strong>of</strong> an interactive simulation model as a<br />
tool for humans to learn about a system and its behaviour.<br />
Research on <strong>the</strong> use <strong>of</strong> computers in instructional settings, or Intelligent Tutoring Systems (ITS), relies<br />
on both <strong>the</strong>ories <strong>of</strong> human learning and automated reasoning. Integration <strong>of</strong> <strong>the</strong>se <strong>the</strong>ories has been an<br />
important context for <strong>the</strong> development <strong>of</strong> systems that are directed to facilitating knowledge<br />
communication between computers and human actors (Wenger, 1987). To transfer knowledge and to<br />
interact with a human being on <strong>the</strong> level <strong>of</strong> knowledge content, an educational system should posses a<br />
dynamic internal representation <strong>of</strong> <strong>the</strong> knowledge (Bouwer et al. 2002; Newell, 1982). Usually <strong>the</strong>se<br />
representations take <strong>the</strong> form <strong>of</strong> models or, more specifically, simulation models. With simulation<br />
models, a central issue with respect to guiding effective knowledge communication is <strong>the</strong> format in<br />
which knowledge can be communicated.<br />
The use <strong>of</strong> numerical (simulation) models to express behavioural features is most eminent in<br />
ma<strong>the</strong>matics and o<strong>the</strong>r quantitatively oriented sciences. <strong>An</strong> alternative way <strong>of</strong> representing behaviour<br />
comes from <strong>the</strong> strand <strong>of</strong> research generally referred to as qualitative reasoning (Forbus, 1984), which<br />
has its origins in <strong>the</strong> field <strong>of</strong> artificial intelligence. Qualitative representations form <strong>the</strong> essence <strong>of</strong> <strong>the</strong><br />
model presented. There are several issues from which qualitative models show <strong>the</strong>ir utility and<br />
usefulness with respect to automated educational settings, see also (Bouwer et al. 2002).<br />
First, when designed to be articulate about <strong>the</strong> knowledge communicated, qualitative models in<br />
educational settings can provide a basis from which guidance to <strong>the</strong> user can be established, for<br />
instance in <strong>the</strong> form <strong>of</strong> generating explanations. Constructing articulate models means that a<br />
simulation model should contain sufficient detail about <strong>the</strong> domain at hand, be explicit about <strong>the</strong> topics<br />
to be learned, and facilitate feedback or explanation where necessary, see also (Falkenhainer &<br />
Forbus, 1992; Winkels & Bredeweg, 1998). Second, in an educational setting, qualitative models<br />
provide a way <strong>of</strong> acquiring knowledge on <strong>the</strong> conceptual level, which is seen as a prerequisite in order<br />
to move on to numerical analysis. Finally, qualitative models can be applied to less formalised<br />
domains in order to identify and depict conceptual (e.g., causal) structures that explain <strong>the</strong> behaviour<br />
<strong>of</strong> a system. This means that qualitative models can be helpful in acquiring specific insights in various<br />
domains.<br />
5
From an educational perspective, interaction with qualitative (simulation) models focuses on <strong>the</strong><br />
notion <strong>of</strong> behaviour analysis. This analysis is based on both prediction and post diction <strong>of</strong> system<br />
behaviour. With qualitative models, <strong>the</strong> tasks <strong>of</strong> prediction and post diction can be associated with<br />
various skills that a learner should master in order to reason about specific (physical) phenomena.<br />
These skills include generation <strong>of</strong> causal accounts, reasoning from structure, and reasoning with<br />
multiple perspectives (Bouwer et al., 2002; Dekleer & Brown, 1984).<br />
Typical domains in qualitative reasoning research are physics problems, such as pressure regulators or<br />
steam engines (Collins & Forbus, 1987). Researchers such as (Guerrin, 1991) and (Heller et al. 1995),<br />
have developed models that are focused on ecological problems. They have shown <strong>the</strong> usefulness <strong>of</strong><br />
using qualitative reasoning in non-physics domains. However, to date <strong>the</strong> amount <strong>of</strong> qualitative<br />
models produced in o<strong>the</strong>r domains than physics is relatively low. There is a need to go beyond <strong>the</strong><br />
scope <strong>of</strong> classical physics problems and show <strong>the</strong> added value <strong>of</strong> qualitative models in non-physical<br />
domains.<br />
Summarising, <strong>the</strong>re is increasing attention being paid to <strong>the</strong> notion <strong>of</strong> communicating domain<br />
knowledge through ITS. Qualitative reasoning is frequently associated to ITS, but it is a framework<br />
for knowledge representation that is traditionally used for reasoning about physics problems. Given<br />
this context <strong>the</strong> research question is formulated as follows:<br />
Design and implement a simulation model which comprises a new qualitative ontology for <strong>the</strong><br />
ecological domain <strong>of</strong> <strong>the</strong> water cycle, which inhibits a knowledge organisation that allows <strong>the</strong><br />
model to function as expert module in an educational setting.<br />
In addition to working out this research question this <strong>the</strong>sis will provide <strong>the</strong>oretical context on<br />
intelligent tutoring systems, qualitative reasoning and constructing simulation models.<br />
The next Chapter provides context on <strong>the</strong> research field <strong>of</strong> ITS. The third chapter introduces and<br />
discusses <strong>the</strong> concept <strong>of</strong> qualitative reasoning. In chapter 4 <strong>the</strong> functioning <strong>of</strong> <strong>the</strong> qualitative simulator<br />
and its knowledge representation is explained. Chapter 5 goes into detail about <strong>the</strong> issues involved<br />
with model construction, in particular qualitative models. The domain model <strong>of</strong> <strong>the</strong> water cycle is<br />
described in chapter 6. In chapter 7 <strong>the</strong> simulation <strong>of</strong> <strong>the</strong> water cycle is discussed. A review <strong>of</strong> <strong>the</strong><br />
models is given in chapter 8. Chapter 9 forms <strong>the</strong> conclusion and discussion <strong>of</strong> <strong>the</strong> research.<br />
6
2. Knowledge Systems in Education: <strong>An</strong> Overview<br />
The context <strong>of</strong> research on Intelligent Tutoring Systems is multidisciplinary. It is linked to <strong>the</strong> fields <strong>of</strong><br />
cognitive psychology (<strong>An</strong>derson, Boyle, Farrell & Reiser, 1987), pedagogy (Rittle-Johnson &<br />
Koedinger, 2001), computer science and subject matter domain <strong>the</strong>ories. In order to communicate<br />
knowledge to students, ITS’s typically rely on several components, such as <strong>the</strong> student model, <strong>the</strong><br />
curriculum planner, <strong>the</strong> simulation engine, <strong>the</strong> user interface and a model <strong>of</strong> <strong>the</strong> subject matter.<br />
Research fields need to collaborate in order to produce a working system out <strong>of</strong> several modules,<br />
which can be considered as expert systems individually.<br />
From <strong>the</strong> beginning <strong>of</strong> research on ITS’s, effort has been put in producing tutorial dialogues,<br />
communication strategies, and in principles for knowledge representation (Wenger 1987). In order to<br />
establish interaction between a computer in <strong>the</strong> role <strong>of</strong> a tutor and a human in <strong>the</strong> role <strong>of</strong> student, one<br />
<strong>of</strong> <strong>the</strong> most fundamental problems is <strong>the</strong> explicit representation <strong>of</strong> <strong>the</strong> knowledge that should be taught<br />
and <strong>the</strong> curriculum for teaching it (Lesgold, 1988). Issues concerning curriculum are generally dealt<br />
with by <strong>the</strong> use <strong>of</strong> <strong>the</strong>ories from pedagogy or cognitive psychology. The research on human<br />
knowledge acquisition and problem solving is a still ongoing topic <strong>of</strong> investigation. Next to research<br />
on <strong>the</strong> use <strong>of</strong> for example procedural knowledge in problem solving, o<strong>the</strong>r questions have risen: How<br />
and to what extent for example are heuristics used in problem solving? What tacit or common sense<br />
knowledge underlies certain problem solving capabilities? One <strong>of</strong> <strong>the</strong> issues is to identify and make<br />
explicit how common sense knowledge is utilised in dealing with physical problems. For example,<br />
Falkenhainer et al., (1986) researched <strong>the</strong> use <strong>of</strong> analogies with subjects when experimenting in<br />
physical domains. A commonly recognised shortcoming <strong>of</strong> <strong>the</strong> older instructional systems was <strong>the</strong><br />
failure to incorporate this common sense knowledge that people use in solving problems.<br />
The question <strong>of</strong> how to represent <strong>the</strong> right kind <strong>of</strong> knowledge has been a source <strong>of</strong> debate since <strong>the</strong><br />
first ITS’s were developed (Wenger, 1987). Early ITS were able to represent domain knowledge by<br />
<strong>the</strong> use <strong>of</strong> frames and slots, which are linear descriptions. The curriculum as a consequence was very<br />
rigid and not sensitive to <strong>the</strong> knowledge <strong>of</strong> students. The systems failed to give articulate accounts for<br />
what <strong>the</strong>y were teaching. <strong>An</strong> alternative to this frame-oriented paradigm was proposed by (Carbonell,<br />
1970) in defining <strong>the</strong> information-structure-oriented paradigm, along with <strong>the</strong> SCHOLAR system, a<br />
geography tutor. The structure <strong>of</strong> knowledge representation used in SCHOLAR is a semantic network<br />
(Quillian, 1968), including relations such as super part, super concepts, et cetera. In terms <strong>of</strong><br />
knowledge communication, Carbonell pioneered on issues that are still relevant in Intelligent<br />
Computer Aided Instruction (ICAI), such as mixed-initiative dialogues, explanation generation and<br />
diagnosis <strong>of</strong> learner behaviour.<br />
In ano<strong>the</strong>r system called WHY, developed by (Collins and Stevens, 1977) special attention was given<br />
to learner diagnosis, more specifically <strong>the</strong> construction <strong>of</strong> Socratic Dialogues, in which functional<br />
7
analysis plays a fundamental role. The knowledge representation in WHY is achieved by using<br />
hierarchical scripts and <strong>the</strong> generation <strong>of</strong> questions is based on <strong>the</strong>se scripts, which actually allowed a<br />
kind <strong>of</strong> evolutionary curriculum on <strong>the</strong> subject. In evaluating <strong>the</strong> system, Collins and Stevens raised<br />
some important issues. Of particular importance was <strong>the</strong> lack <strong>of</strong> incorporating high level goals in<br />
tutoring, a result <strong>of</strong> only local applicability <strong>of</strong> question generating rules. This led to <strong>the</strong> appraisal <strong>of</strong> <strong>the</strong><br />
role that causal models and procedural knowledge play in pursuing high level goals, next to just<br />
declarative knowledge. The use <strong>of</strong> subgoals and backtracking are common examples <strong>of</strong> procedural<br />
knowledge in resolving problems, and should thus be incorporated in instructional systems.<br />
The use <strong>of</strong> artificial intelligence in tutoring systems has shifted <strong>the</strong> focus form encoding decisions to<br />
<strong>the</strong> encoding <strong>of</strong> knowledge and how knowledge can be communicated (Wenger, 1987). <strong>An</strong> important<br />
contribution from <strong>the</strong> A.I. community was made by <strong>the</strong> investigation on how certain types <strong>of</strong><br />
knowledge (e.g. procedural, causal) may be encoded in generic form. When considering most domain<br />
knowledge, different views may focus on different aspects <strong>of</strong> a system, and various levels <strong>of</strong> detail<br />
may be required for particular views. A more dynamic kind <strong>of</strong> encoding has a one to many mapping,<br />
and makes use <strong>of</strong> explicit knowledge about <strong>the</strong> applicability <strong>of</strong> certain knowledge. The idea <strong>of</strong><br />
multiple viewpoints on a system relates to <strong>the</strong> notion <strong>of</strong> a mental model that is able to simulate <strong>the</strong><br />
behaviour <strong>of</strong> (physical) systems to a certain extent. <strong>An</strong> important feature <strong>of</strong> <strong>the</strong>se models is that <strong>the</strong>y<br />
usually incorporate multiple viewpoints on <strong>the</strong> same subject. From a computational point <strong>of</strong> view, this<br />
is called <strong>the</strong> cognitive modelling approach.<br />
Work on ITS’s focuses on development <strong>of</strong> simulations that mimic human problem solving, including<br />
humanlike deployment <strong>of</strong> knowledge (<strong>An</strong>derson, 1993). The model <strong>the</strong>refore requires reasoning<br />
abilities about it's own content. Research on mental models, as used by people in reasoning tasks,<br />
supported <strong>the</strong> work on computational models. The idea here is that people conceive <strong>of</strong> models in<br />
different scopes e.g. planetary climates water cycle etc., and <strong>the</strong>n an approach could be to<br />
develop a view <strong>of</strong> <strong>the</strong> mental model with models that are decomposed at <strong>the</strong> functional, aggregate and<br />
molecular level, which can help explain each o<strong>the</strong>r. <strong>An</strong>d indeed, <strong>the</strong>se ideas have proven to be<br />
essential to questions about sequencing <strong>the</strong> knowledge to be conveyed to learner i.e., to resemble <strong>the</strong><br />
way a teacher would plan a lecture when trying to explain a subject, successively addressing deeper<br />
levels and higher complexity (Collins & Gentner 1983).<br />
<strong>An</strong> important task emerged from <strong>the</strong> standpoint <strong>of</strong> computational models, namely a formalised<br />
framework being able to represent <strong>the</strong> knowledge that people use in reasoning about dynamic physical<br />
phenomena. The framework should be able to derive <strong>the</strong> same general conclusions about physical<br />
problems as ma<strong>the</strong>matical models, albeit in a less complex fashion, from an algebraic point <strong>of</strong> view.<br />
This trajectory <strong>of</strong> research that was established came to be known under several different names, <strong>the</strong><br />
most common are Naive Physics and Qualitative Reasoning. As for Naive Physics Patrick Hayes<br />
(1985), introduced <strong>the</strong> topic most seriously in <strong>the</strong> field <strong>of</strong> A.I. and expert systems. He expressed <strong>the</strong><br />
need to abandon ‘toy world problems’ in favour <strong>of</strong> more complex formal models <strong>of</strong> <strong>the</strong> world as it is<br />
8
intuitively perceived by us, in terms <strong>of</strong> objects, change, events and processes. Among <strong>the</strong> motivations<br />
for developing qualitative models lies, much <strong>the</strong> same as with WHY, <strong>the</strong> desire to communicate causal<br />
knowledge to learners. The SOPHIE project (Brown et al., 1982) serves as a good example. For<br />
Qualitative Reasoning, <strong>the</strong> focus is on implemented models containing causal knowledge needed to<br />
reason about physical processes, and among <strong>the</strong> initiating efforts to do so are (Forbus 1984, Kuipers<br />
1986). The potential for qualitative simulation models in tutoring paradigms is a research topic itself<br />
(<strong>An</strong>derson 1993). The most clear potential lies in terms <strong>of</strong> explanation, curriculum sequencing (see<br />
also White & Frederiksen, 1986; Salles & Bredeweg, 2001) and <strong>the</strong> generation <strong>of</strong> articulate<br />
simulations <strong>of</strong> specific systems. When we want subjects to learn about how to interact with a physical<br />
system by means <strong>of</strong> a tutoring system, it is necessary to mimic human mental models <strong>of</strong> systems.<br />
Qualitative representations are well suited to communicate causal knowledge and are able to support<br />
<strong>the</strong> notion <strong>of</strong> curriculum sequencing multiple perspectives. By giving a model an explicit purpose we<br />
should be able to limit <strong>the</strong> amount <strong>of</strong> inferences and concepts needed to create a sufficiently<br />
sophisticated model for <strong>the</strong> (instructional) purpose or goal at hand.<br />
The research on ITS is nowadays directed to <strong>the</strong> development <strong>of</strong> formal systems that can perform<br />
humanlike reasoning about systems and at <strong>the</strong> same time be explicit about <strong>the</strong> reasoning steps it<br />
performs. <strong>An</strong> important issue forms <strong>the</strong> models being articulate about <strong>the</strong> knowledge, which is<br />
facilitated by developments in qualitative physics (Forbus et al. 2000). <strong>An</strong> example <strong>of</strong> such an<br />
articulate learning environment, or virtual laboratory, is <strong>Cycle</strong>Pad (Forbus & Whalley, 1994) where<br />
<strong>the</strong>rmodynamic cycles can be analysed.<br />
Many different types <strong>of</strong> tutoring systems have been developed. At this stage, <strong>the</strong> research has not only<br />
benefited from <strong>the</strong> diverse <strong>of</strong>fer in tutoring systems; lots <strong>of</strong> systems have been developed<br />
independently and at this moment <strong>the</strong> sharing <strong>of</strong> knowledge is problematic. As a reaction, effort has<br />
been put in <strong>the</strong> development <strong>of</strong> frameworks for interoperable and component-based systems (Ritter &<br />
Koedinger, 1997; Koedinger, Su<strong>the</strong>rs & Forbus, 1998). Next to easier maintainance and easier re-use<br />
<strong>of</strong> components, <strong>the</strong> functionalities can be constructed by composition <strong>of</strong> components such as<br />
simulation models, learning agenda’s. <strong>An</strong> important requirement that follows from <strong>the</strong> above is <strong>the</strong><br />
need for a shared semantic between different components in a learning environment, so that<br />
components can communicate and share information smoothly.<br />
9
3. Qualitative Reasoning Paradigm<br />
Qualitative reasoning is now a well-established field within artificial intelligence. At this time a whole<br />
range <strong>of</strong> approaches and methods within qualitative reasoning have emerged. Among its goals is a<br />
qualitative computational framework for reasoning about physics. This means as much as rephrasing<br />
traditional physics in an alternative way, while preserving its laws and observations. Several<br />
motivations form <strong>the</strong> basis <strong>of</strong> research on this alternative framework, (Forbus, 1988). One <strong>of</strong> <strong>the</strong><br />
original motivations came from <strong>the</strong> research on cognitive models. Here efforts have been done to<br />
create computational accounts for <strong>the</strong> mental models (Gentner & Stevens, 1983) that people have<br />
when interacting in complex physical systems (White & Frederiksen 1990; Dekleer & Brown, 1984).<br />
The mental models that people use, are conceived <strong>of</strong> as runnable which means that some kind <strong>of</strong><br />
simulation is involved. This simulation is sometimes explained as qualitative simulation, involving<br />
general common sense inferences.<br />
<strong>An</strong>o<strong>the</strong>r motivation comes from <strong>the</strong> field <strong>of</strong> robotics, where <strong>the</strong>se ideas are put to practice in order to<br />
allow robots to move in <strong>the</strong>ir physical environment (Hayes, 1978; Zhao, 1995). When acting in <strong>the</strong><br />
physical world, <strong>the</strong> plain use <strong>of</strong> a set <strong>of</strong> equations to analyse and act on <strong>the</strong>ir environment is not<br />
sufficient to deal with complex surroundings. Equations in <strong>the</strong>mselves contain no information about<br />
<strong>the</strong>ir applicability in physical situations. Robots have to rely on so called tacit or common sense<br />
knowledge in order to determine what kind <strong>of</strong> knowledge is applicable to a certain situation, and to<br />
mentally simulate and reason about dynamic processes.<br />
A third motivation for qualitative physics research lies in <strong>the</strong> field already discussed in section two,<br />
namely research that is directed to knowledge communication for educational purposes. Qualitative<br />
physics provides an articulate vocabulary that corresponds to people’s tacit conceptions <strong>of</strong><br />
phenomena. Qualitative knowledge is usually modelled in terms <strong>of</strong> intuitive components like causal<br />
relations, processes, etc. As such <strong>the</strong>y provide a basis to facilitate knowledge communication to<br />
learners through knowledge intensive educational systems. Qualitative models form a basis for<br />
developing dialogs and o<strong>the</strong>r forms <strong>of</strong> interaction, such as explanation generation (Wenger, 1987;<br />
Aleven & Koedinger, 2000).<br />
The idea to develop an alternative physics stems from <strong>the</strong> fact that traditional physics lacks to produce<br />
a causal account <strong>of</strong> phenomena, something that is a core feature <strong>of</strong> mental models as deployed by<br />
humans. Although this looks relatively straightforward <strong>the</strong> question <strong>of</strong> how for example to represent<br />
causal ordering remains widely discussed within <strong>the</strong> QR community. When modelling a domain,<br />
choices such as ‘X causes Y, and not <strong>the</strong> o<strong>the</strong>r way around’ are sometimes difficult to prove, and<br />
<strong>the</strong>refore several representations <strong>of</strong> causality have been proposed (Forbus & Gentner 1986, Iwasaki &<br />
Simon 1986). It is generally agreed to that traditional physics or ma<strong>the</strong>matics do a good job when it<br />
comes to generating precise descriptions or simulations <strong>of</strong> physical situations. However, a very<br />
10
accurate set <strong>of</strong> input values that apply to physical equations is required. When no precise information<br />
is available, <strong>the</strong> use <strong>of</strong> numerical simulations tends to become computationally expensive, as large<br />
ranges <strong>of</strong> values must be simulated. In qualitative physics methods exist for representing partial<br />
information about numerical values, abandoning traditional use <strong>of</strong> numbers. These methods include<br />
<strong>the</strong> use <strong>of</strong> signs and ordinals (Forbus, 1984). Fur<strong>the</strong>rmore, traditional descriptions based on incomplete<br />
information gives no guarantee that all possible behaviours will be found, and <strong>the</strong> very availability <strong>of</strong> a<br />
set <strong>of</strong> (exhaustive) alternatives is required in many reasoning problems. Typical to qualitative models<br />
is that an exhaustive set <strong>of</strong> alternative behaviours can be generated, in particular under circumstances<br />
with incomplete information. The formulation <strong>of</strong> explicit modelling assumptions (Falkenhainer &<br />
Forbus, 1991) under which a model is applicable provides <strong>the</strong> means to specify <strong>the</strong> boundaries <strong>of</strong> <strong>the</strong><br />
analysis, and <strong>the</strong>reby automatically controls <strong>the</strong> amount <strong>of</strong> possible behaviours found. Between <strong>the</strong>se<br />
boundaries <strong>the</strong> analysis is exhaustive.<br />
<strong>An</strong>o<strong>the</strong>r feature <strong>of</strong> <strong>the</strong> alternative physics through qualitative reasoning is reasoning from structure<br />
(Dekleer, 1984). Physical phenomena are represented in terms <strong>of</strong> local descriptions <strong>of</strong> objects and<br />
processes that can be combined to form complex structures. Behaviour is derived from <strong>the</strong><br />
interconnection <strong>of</strong> different objects and processes. Most qualitative models rely on a blend <strong>of</strong> both<br />
generic components and domain specific components. The development <strong>of</strong> generic knowledge allows<br />
models to have wide coverage and multi-purpose, so that <strong>the</strong> inferences captured by qualitative models<br />
are not particular to certain instances <strong>of</strong> a physical situation, but to classes <strong>of</strong> similar situations. People<br />
in general do use some general principles when coping with <strong>the</strong> environment, however, <strong>the</strong> link<br />
between human mental reasoning and <strong>the</strong> use <strong>of</strong> generic knowledge has not been shown directly. It<br />
seems likely that people use some domain-independent laws and principles when reasoning about <strong>the</strong><br />
physical world (Forbus & Gentner, 1997). The modularity <strong>of</strong> <strong>the</strong> knowledge in itself allows for<br />
composition <strong>the</strong>ories, which account for a high degree <strong>of</strong> reusability <strong>of</strong> <strong>the</strong> knowledge. Components <strong>of</strong><br />
knowledge in qualitative physics will be referred to as model fragments subsequently.<br />
The above has sketched some general motivations and purposes <strong>of</strong> qualitative physics as well as<br />
concepts that are at <strong>the</strong> basis <strong>of</strong> qualitative physics. In general <strong>the</strong>y are concerned with formalising<br />
aspects that constitute <strong>the</strong> mental processing involved with reasoning about <strong>the</strong> physical world. The<br />
next section will outline some general methods <strong>of</strong> qualitative reasoning.<br />
3.1 Describing change<br />
The representation <strong>of</strong> dynamics within qualitative physics itself varies considerably. Generally<br />
speaking, (qualitative) representations <strong>of</strong> dynamics involve numbers, equations and functions. One <strong>of</strong><br />
<strong>the</strong> most influential <strong>the</strong>ories on qualitative dynamics is <strong>the</strong> Qualitative Process Theory (Forbus, 1984),<br />
which comprises a qualitative ontology for <strong>the</strong> representation <strong>of</strong> physical processes. A technical<br />
11
discussion on <strong>the</strong> representational primitives used in this research will be provided when discussing<br />
<strong>the</strong> knowledge and reasoning framework used for this research, see chapter 4. Some general features<br />
<strong>of</strong> dynamic representations are discussed next.<br />
Quantities in qualitative reasoning have an amount and a derivative which both have signs. The most<br />
widely used method <strong>of</strong> reasoning with changing quantities used in QR is <strong>the</strong> sign algebra, which uses<br />
measurements like minus, plus or zero for <strong>the</strong> derivative <strong>of</strong> a parameter. By using sign algebra for <strong>the</strong><br />
representation <strong>of</strong> ordinary numbers in qualitative models, precise predictions cannot be made<br />
anymore, but <strong>the</strong>y allow abstract descriptions <strong>of</strong> behaviour. The representation <strong>of</strong> a set <strong>of</strong> continuous<br />
variables that a quantity may have is done with a partial ordering. A partial ordering contains a set <strong>of</strong><br />
ordinal values, with zero or chosen landmarks as points <strong>of</strong> comparison. This type <strong>of</strong> representation is<br />
called a quantity space. Within <strong>the</strong> partial ordering, <strong>the</strong> values with no values between <strong>the</strong>m are called<br />
neighbours. Between neighbours intervals exist, which have no defined length. A quantity space is<br />
unique for <strong>the</strong> quantity it is assigned to.<br />
In order to describe quantities and <strong>the</strong> values that <strong>the</strong>y may have, inequalities are used. This way,<br />
values <strong>of</strong> quantities may be compared to each o<strong>the</strong>r or to certain limit points that may exist, such as <strong>the</strong><br />
boiling point <strong>of</strong> a certain liquid. While each quantity space is unique, we need to be able to express<br />
that some values <strong>of</strong> quantities correspond. It is possible by defining for example that for a particular<br />
quantity ‘volume’, <strong>the</strong> values low, medium and high can be set equal to <strong>the</strong> value positive for <strong>the</strong><br />
quantity ‘pressure’.<br />
In order to induce change, <strong>the</strong> use <strong>of</strong> functions and equations is required. The equations that are used<br />
in most qualitative reasoning concern straightforward arithmetic equations. The functions that are used<br />
most in qualitative reasoning are monotonic functions. A correspondence for example is a function<br />
that derives <strong>the</strong> value <strong>of</strong> a parameter from <strong>the</strong> value <strong>of</strong> <strong>the</strong> parameter that it corresponds to.<br />
Qualitative proportionalities are used to relate <strong>the</strong> derivative <strong>of</strong> parameters. When processes are being<br />
modelled, <strong>the</strong>ir effect are specified using influences. Influences affect <strong>the</strong> derivative <strong>of</strong> a parameter for<br />
a particular number <strong>of</strong> states. This type <strong>of</strong> influence is independent on changes in o<strong>the</strong>r quantities and<br />
are <strong>the</strong>refore called direct influences. O<strong>the</strong>r influences depend on changes in a related quantity and are<br />
<strong>the</strong>refore called indirect influences.<br />
3.2 Compositional <strong>Model</strong>ling<br />
One <strong>of</strong> <strong>the</strong> desiderata for a <strong>the</strong>ory <strong>of</strong> qualitative dynamics is composability <strong>of</strong> system models, which<br />
aims to comply with <strong>the</strong> no-function-in-structure principle (DeKleer & Brown, 1984). This principle<br />
states that partial models do not assume <strong>the</strong> functioning <strong>of</strong> <strong>the</strong> system as a whole. In modelling<br />
dynamic systems, a now common strategy is <strong>the</strong> compositional modelling approach by (Falkenhainer<br />
& Forbus 1991), where models can be generated by assembling several model fragments <strong>of</strong> domain<br />
<strong>the</strong>ories. The main benefits <strong>of</strong> this modular-based development <strong>of</strong> models are <strong>the</strong> possibilities <strong>of</strong> reuse<br />
<strong>of</strong> components, easier maintenance and extension <strong>of</strong> <strong>the</strong> models. All model fragments contain a<br />
12
condition part and a given part. <strong>An</strong> assembly <strong>of</strong> model fragments is instantiated on <strong>the</strong> basis <strong>of</strong> an<br />
initial scenario, which contains knowledge <strong>of</strong> <strong>the</strong> model fragments <strong>of</strong> interest. When <strong>the</strong> elements in<br />
<strong>the</strong> scenario correspond to <strong>the</strong> condition part <strong>of</strong> a particular model fragment, it becomes instantiated.<br />
The elements in a scenario may match with multiple partial models. A scenario typically activates<br />
multiple partial models, which toge<strong>the</strong>r form a system description at a particular time. Decomposition<br />
<strong>of</strong> <strong>the</strong> domain into separate components provides <strong>the</strong> conditions for approaches to automated<br />
modelling <strong>of</strong> a domain. Automated modelling constructs <strong>the</strong> most efficient model <strong>of</strong> a particular<br />
scenario, by controlling <strong>the</strong> amount <strong>of</strong> instantiated knowledge. Composability <strong>of</strong> <strong>the</strong> model<br />
components has <strong>the</strong> advantage that not every possible scenario has to be anticipated.<br />
The domain <strong>the</strong>ory itself is written down in model fragments, which can be combined to form<br />
descriptions <strong>of</strong> particular physical phenomena, given an initial scenario description. With respect to<br />
<strong>the</strong> applicability <strong>of</strong> model fragments in some models, modelling assumptions can be defined. In<br />
general, modelling assumptions constrain <strong>the</strong> amount <strong>of</strong> detail considered in an analysis. Simplifying<br />
assumptions can be used to control <strong>the</strong> set <strong>of</strong> model fragments that will be instantiated, and <strong>the</strong>refore<br />
can provide global structure in <strong>the</strong> model by specifying and controlling ontology, grain size or<br />
perspective <strong>of</strong> a model under consideration. <strong>An</strong>o<strong>the</strong>r type <strong>of</strong> modelling assumption is operating<br />
assumptions, which describe <strong>the</strong> kinds <strong>of</strong> behaviour relevant to a task, such as a steady state<br />
assumption. The main advantage <strong>of</strong> using <strong>of</strong> modelling assumptions is that incompatible or competing<br />
views on individuals may exist side by side in a library <strong>of</strong> model fragments. <strong>Model</strong>ling assumptions<br />
are written down in <strong>the</strong> format <strong>of</strong> model fragments. In logical form, a model fragment is encoded as a<br />
first order implication (Falkenhainer & Forbus, 1991 p.103):<br />
I ∧A ∧O R,<br />
where<br />
• I is a set <strong>of</strong> structural conditions.<br />
• A is a possibly empty set <strong>of</strong> assumptions concerning <strong>the</strong> fragments relevance to questions <strong>of</strong><br />
interest.<br />
• O is a possibly empty set <strong>of</strong> operating conditions including inequalities between parameters,<br />
conditions on activity <strong>of</strong> o<strong>the</strong>r model fragments and operating assumptions.<br />
• R is a set <strong>of</strong> relations imposed by <strong>the</strong> model fragment.<br />
13
3.3 Qualitative simulation<br />
The tasks that people perform in <strong>the</strong>ir environment <strong>of</strong>ten depend on <strong>the</strong> notion <strong>of</strong> behaviour analysis,<br />
which involves, among o<strong>the</strong>rs, predictive or postdictive reasoning steps. A problem with traditional<br />
physics is that <strong>the</strong>re is no means to show how a particular state <strong>of</strong> affairs has come about, and which<br />
trajectory or history is tied to a certain parameter. Qualitative simulation produces a set <strong>of</strong> qualitative<br />
states and <strong>the</strong> transitions between <strong>the</strong>m. The result is an explicit causal account <strong>of</strong> events. Qualitative<br />
states are used to represent change over time in a particular system. When nothing changes, than <strong>the</strong>re<br />
will be no new qualitative state. A state thus is a qualitative description <strong>of</strong> a system at a particular<br />
moment. In order to perform behaviour analysis, it is necessary to identify <strong>the</strong> various actors that are<br />
involved in <strong>the</strong> environment and how <strong>the</strong>y interact over spans <strong>of</strong> time. Conversely, <strong>the</strong> question <strong>of</strong><br />
what can be assumed to have no effect is equally important. In terms <strong>of</strong> QR, what to include in an<br />
analysis and what to leave out is captured in <strong>the</strong> frame problem (McCarthy 1963; McCarthy & Hayes<br />
1969). Patrick Hayes introduced <strong>the</strong> notion <strong>of</strong> histories into <strong>the</strong> description <strong>of</strong> events (Hayes, 1985).<br />
Histories describe events in a temporally unbounded and spatially bounded way. The result is that<br />
events can be analysed locally over time, which makes it possible to examine <strong>the</strong> causality <strong>of</strong> subparts<br />
<strong>of</strong> systems.<br />
Using QR, <strong>the</strong> causal account <strong>of</strong> <strong>the</strong> behaviour is one <strong>of</strong> <strong>the</strong> main interests <strong>of</strong> building <strong>the</strong> models. In a<br />
qualitative simulation, a causal path is usually generated by a set <strong>of</strong> possible state transitions form a<br />
given input. This kind <strong>of</strong> simulation allows to handle incomplete information, or information with low<br />
resolution, which is typical with common sense reasoning. The collection <strong>of</strong> all possible state<br />
transitions is called a qualitative envisionment (DeKleer 1977). A qualitative envisionment should<br />
retain <strong>the</strong> same conclusions about behaviour as <strong>the</strong> actual behaviour found in <strong>the</strong> referent, <strong>the</strong> real<br />
world system. The relation between histories and envisionment remains debated in <strong>the</strong> QR<br />
community. Generally speaking, one can say that each history resembles a path through an<br />
envisionment but not <strong>the</strong> o<strong>the</strong>r way around (Forbus, 1988).<br />
3.4 Ontology<br />
With <strong>the</strong> idea <strong>of</strong> a broadly applicable qualitative physics that was discussed earlier, ontologies<br />
containing useful and common abstractions are <strong>of</strong> great utility. The task <strong>of</strong> creating an ontology<br />
capable <strong>of</strong> representing qualitative physics is a central <strong>the</strong>me in <strong>the</strong> field <strong>of</strong> qualitative reasoning. We<br />
have seen that with qualitative kinematics different worldviews may exist on how to model<br />
interconnections between components. In a more fundamental sense, three <strong>of</strong> <strong>the</strong> most influential<br />
approaches within qualitative modelling have become <strong>the</strong> component-based approach (Brown & de<br />
Kleer, 1984), <strong>the</strong> process centred approach (Forbus, 1984) and <strong>the</strong> constrained based approach<br />
(Kuipers, 1986). Although each <strong>of</strong> <strong>the</strong> approaches has its own merits, for a detailed comparison<br />
14
etween <strong>the</strong> methods <strong>the</strong> reader is referred to (Bredeweg ,1992). Salles et al. (1996) compared <strong>the</strong><br />
approaches on <strong>the</strong>ir individual suitability for modelling ecological systems.<br />
When modelling dynamic systems qualitatively, various commitments have to be made about how <strong>the</strong><br />
world is represented, and thus which qualitative ontology will be used. Collins and Forbus (1989) for<br />
example, show that <strong>the</strong> process-centred ontology developed by Kenneth Forbus (1984) in <strong>the</strong><br />
Qualitative Process Theory (QPT), adopts well to <strong>the</strong> field <strong>of</strong> <strong>the</strong>rmodynamics, where most <strong>of</strong> <strong>the</strong><br />
behaviour can be interpreted in <strong>the</strong> sense <strong>of</strong> processes that produce heat flows, energy exchanges, or<br />
work. <strong>An</strong>o<strong>the</strong>r modelling ontology used in qualitative physics is <strong>the</strong> device centred ontology<br />
advocated by (De Kleer & Brown, 1984). They have adopted <strong>the</strong> approach with <strong>the</strong> modelling <strong>of</strong><br />
electric circuits. Here <strong>the</strong> system is divided into components that have a set <strong>of</strong> equations related to<br />
<strong>the</strong>m, and a network <strong>of</strong> components can be analysed by for example a constraint propagation method.<br />
By using local descriptions <strong>of</strong> components it should be avoided that <strong>the</strong> behaviour <strong>of</strong> components<br />
assume <strong>the</strong> functioning <strong>of</strong> a device as a whole, ra<strong>the</strong>r than explaining <strong>the</strong> local interaction between<br />
components. When considering ecological or biological domains, see for example (Guerrin, 1990), yet<br />
again a different ontology can be used. Of special relevance in <strong>the</strong>se domains are phenomena that have<br />
to be modelled in terms <strong>of</strong> orders <strong>of</strong> magnitude. Orders <strong>of</strong> magnitude can be used when it is needed to<br />
say that for example one process is submissive to ano<strong>the</strong>r process and need not be considered in <strong>the</strong><br />
outcome. For example <strong>the</strong> amount <strong>of</strong> rain droplets that evaporate when falling towards <strong>the</strong> earth can be<br />
neglected compared to <strong>the</strong> amount that reaches <strong>the</strong> earth.<br />
Fur<strong>the</strong>rmore, different levels <strong>of</strong> granularity may be appropriate depending on <strong>the</strong> purpose <strong>of</strong> <strong>the</strong><br />
model. For example <strong>the</strong> representation <strong>of</strong> liquids in relation to <strong>the</strong> objects containing <strong>the</strong>m alone has<br />
led to representations at <strong>the</strong> molecular level, e.g. <strong>the</strong> piece <strong>of</strong> stuff ontology, <strong>the</strong> contained stuff<br />
ontology (Hayes, 1979) and more recently <strong>the</strong> bounded stuff ontology (Kim, 1993). In our work we<br />
will introduce a particular configuration <strong>of</strong> contained liquids, which deals with vertical differences<br />
between contained liquids. Our ontology <strong>the</strong>refore is based on <strong>the</strong> general notions provided by Hayes´<br />
ontology for liquids.<br />
With a wide field <strong>of</strong> application area’s, no method is sufficient to capture all sorts <strong>of</strong> systems in which<br />
qualitative physics is used. As Forbus (1988) puts it, combinations <strong>of</strong> methods may be made given<br />
particular domain modelling purposes.<br />
15
4. Garp: a specific simulator<br />
This section discusses qualitative primitives and <strong>the</strong> way reasoning is performed within a specific<br />
qualitative reasoning approach. Therefore <strong>the</strong> general architecture for reasoning about physics<br />
(GARP), will be introduced. GARP integrates <strong>the</strong> various approaches on <strong>the</strong> representation <strong>of</strong><br />
knowledge <strong>of</strong> dynamic systems. The reasoning engine has been recently extended with a visual<br />
simulation inspection tool, VisiGarp and for this <strong>the</strong>sis, also a graphical model building environment,<br />
HOMER has been used that produces models that can be executed by GARP.<br />
4.1 GARP<br />
As a framework for qualitative knowledge representation, GARP (Bredeweg, 1992) interprets and<br />
simulates qualitative descriptions <strong>of</strong> real world domains. In <strong>the</strong> framework <strong>the</strong> general qualitative<br />
vocabulary is implemented using PROLOG. It supports <strong>the</strong> use <strong>of</strong> model fragments and <strong>the</strong> definition<br />
<strong>of</strong> scenarios, which makes it useful for adopting a compositional modelling approach. The<br />
organisation <strong>of</strong> knowledge in GARP involves several types <strong>of</strong> knowledge:<br />
• The static description <strong>of</strong> <strong>the</strong> actual entities involved in <strong>the</strong> system: <strong>the</strong> system elements in GARP.<br />
The entities are specified in an isa-hierarchy, and structural relations may be declared by using<br />
has-attribute statements<br />
• Primitives that can represent behaviour: parameters, values and dependencies in GARP,<br />
• Specification <strong>of</strong> <strong>the</strong> behaviours that may occur: in terms <strong>of</strong> GARP <strong>the</strong>y are called partial behaviour<br />
models (model fragments), scenario’s (input systems) and transformation rules.<br />
The remainder <strong>of</strong> this section illustrates <strong>the</strong> qualitative statements that are used in <strong>the</strong> GARP<br />
environment. In <strong>the</strong> last section, <strong>the</strong> concept <strong>of</strong> model fragments and <strong>the</strong>ir role in <strong>the</strong> framework are<br />
revisited.<br />
4.2 Quantities<br />
The most widely used method <strong>of</strong> reasoning with quantities used in QR is <strong>the</strong> sign algebra, which uses<br />
measurements like zero, plus or minus for <strong>the</strong> derivative <strong>of</strong> a parameter. Quantities have an amount<br />
and a derivative which both have signs. The representation <strong>of</strong> <strong>the</strong> continuous set <strong>of</strong> variables that a<br />
parameter may have is done with a partial ordering, using point and intervals. Each quantity has it’s<br />
own set <strong>of</strong> values, which doesn’t mean that for some parameters <strong>the</strong> values may correspond.<br />
Quantities are declared as follows:<br />
16
pressure( instance_name, Pressure, _, zp )<br />
The first predicate name specifies <strong>the</strong> parameter type, <strong>the</strong> second argument a variable to which <strong>the</strong><br />
pressure is assigned. The third argument is <strong>the</strong> unique name for <strong>the</strong> combination <strong>of</strong> <strong>the</strong> parameter and<br />
<strong>the</strong> variable. The last argument represents <strong>the</strong> quantity space that is assigned to <strong>the</strong> quantity.<br />
Whenever a value <strong>of</strong> a quantity must be specified, a statement is made which sets <strong>the</strong> value <strong>of</strong> a<br />
quantity to a particular point. Also <strong>the</strong> qualitative derivative can be set in <strong>the</strong> same statement. The<br />
value and derivative can be described as follows:<br />
value( quantity_instance, _, value, derivative )<br />
Value statements like this are <strong>of</strong>ten used for scenario's, in which a particular status <strong>of</strong> <strong>the</strong> system may<br />
be set. The dynamic representation <strong>of</strong> behaviour is done by <strong>the</strong> introducing quantities that affect <strong>the</strong><br />
objects and entities <strong>of</strong> <strong>the</strong> system over time. This change is <strong>the</strong> consequence <strong>of</strong> a shift in <strong>the</strong><br />
(qualitative) value or derivative <strong>of</strong> a parameter. The direction <strong>of</strong> change is stated through <strong>the</strong><br />
derivative (min, zero, plus) <strong>of</strong> <strong>the</strong> value <strong>of</strong> <strong>the</strong> influenced quantity.<br />
Two or more parameters can be dependent on each o<strong>the</strong>r or constrain each o<strong>the</strong>r through various<br />
functional relationships, called quantity relations. <strong>An</strong> exhaustive listing and discussion <strong>of</strong> <strong>the</strong> possible<br />
dependencies between parameters can be found in (Bredeweg, 1992). The following will give <strong>the</strong><br />
reader a short overview <strong>of</strong> <strong>the</strong> available relations.<br />
4.3 Equality<br />
The equality function allows quantities to be set to smaller, equal or greater than a referent. Quantities<br />
may be set equal to a value or two parameters may be set equal, <strong>the</strong> latter meaning that <strong>the</strong> qualitative<br />
values are equal:<br />
equal( Temperature, zero )<br />
smaller( Temperature, Heat )<br />
A Quantity may also be assigned a particular derivative:<br />
d_equal( Temperature, plus )<br />
17
Equality statements are also used to express equations involving two quantities. The equations concern<br />
arithmetic operations such as adding and subtraction.<br />
equal( Pressure_difference, min( PressureA, PressureB ) )<br />
This equation will result in a qualitative value for <strong>the</strong> quantity ‘pressure difference’ as ‘Pressure B’ is<br />
being subtracted from ‘Pressure A’. In <strong>the</strong> case that both ‘Pressure A’ and ‘Pressure B’ have equal<br />
qualitative values but ‘Pressure B’ is increasing, <strong>the</strong> equation will assign a positive value to ‘pressure<br />
difference’.<br />
Two points may be equal but belong to different quantity spaces. This means that a point A from<br />
quantity space 1 is equal to point B <strong>of</strong> quantity space 2. These points can be set equal by using a value<br />
statement <strong>of</strong> <strong>the</strong> type:<br />
equal( top( LevelA ), bottom( LevelB ) )<br />
4.4 Functional relationships<br />
In a similar way, when two quantities have different quantity spaces, <strong>the</strong>ir values can be related as<br />
follows<br />
dir_value_correspondence( Pressure, plus, Level, max )<br />
The relation is directed, which means that <strong>the</strong> value <strong>of</strong> ‘Pressure’ will be set to plus whenever ‘Level’<br />
reaches max. When two quantities share <strong>the</strong> same set <strong>of</strong> values a relation <strong>of</strong> <strong>the</strong> type quantity space<br />
correspondence may be defined:<br />
dir_q_correspondence( Pressure, Volume )<br />
The relation given here is a directed quantity correspondence. which means that <strong>the</strong> value <strong>of</strong> ‘Pressure’<br />
can only be determined if <strong>the</strong> corresponding value <strong>of</strong> ‘Volume’ is known. A similar relation is an<br />
undirected value correspondence:<br />
q_correspondence( Pressure, Volume )<br />
18
It will result in <strong>the</strong> pressure and volume obtaining <strong>the</strong> same qualitative value each subsequent state.<br />
Not only <strong>the</strong> values <strong>of</strong> quantities may be related; <strong>the</strong> signs <strong>of</strong> derivatives can be functionally related as<br />
well.<br />
prop_pos( Level_liquid, Amount_<strong>of</strong>_liquid )<br />
This proportionality means that Level_liquid is increasingly monotonic in its dependence on ‘Amount<br />
<strong>of</strong> liquid’. The inverse <strong>of</strong> this function can also be specified.<br />
prop_neg( Pressure, Outflow )<br />
This is a reverse proportionality, where an increase or positive sign for ‘Outflow’ leads to a decrease,<br />
or negative derivative for ‘Pressure’.<br />
When processes are active, <strong>the</strong>ir effects are usually specified by direct influences. These relations<br />
usually form <strong>the</strong> basis <strong>of</strong> <strong>the</strong> propagation chain.<br />
inf_neg_by( Liquid_source, Flow_rate )<br />
The statement means that <strong>the</strong> value <strong>of</strong> ‘Flow_rate’ has an opposite effect on <strong>the</strong> derivative <strong>of</strong> ‘Liquid<br />
source’<br />
4.5 <strong>Model</strong> Fragments<br />
The partial models as defined in GARP that must contain <strong>the</strong> knowledge about <strong>the</strong> behaviour to be<br />
predicted are comparable to <strong>the</strong> notion <strong>of</strong> model fragments as defined by (Falkenhainer & Forbus<br />
1991), and will be referred to as such subsequently. <strong>Model</strong> fragments are descriptions <strong>of</strong> components<br />
<strong>of</strong> <strong>the</strong> system and may be considered as units that can be combined to describe a greater part <strong>of</strong> <strong>the</strong><br />
system. <strong>Model</strong> fragments may describe views or processes and consist <strong>of</strong> conditions and actions. For a<br />
model fragment to become instantiated <strong>the</strong> conditions that may consist <strong>of</strong> system elements, system<br />
structures, parameters, parameter relations, parameter values and supertype relations must hold;<br />
• System elements are <strong>the</strong> descriptions <strong>of</strong> <strong>the</strong> static elements involved; generally objects and<br />
structural relations.<br />
• System structures refer to o<strong>the</strong>r partial models that must be instantiated<br />
• Quantities are <strong>the</strong> quantifications that belong to system elements<br />
• Quantity relations specify <strong>the</strong> relations that exist for or between parameters in <strong>the</strong> description<br />
19
• Quantity values state <strong>the</strong> exact value and derivative that must hold<br />
• Supertype relations are named when <strong>the</strong> partial model is a subtype <strong>of</strong> ano<strong>the</strong>r partial model<br />
The action, or givens, that a model fragment will give as output is expressed in terms <strong>of</strong> system<br />
elements and parameter relations that are to be instantiated. This is almost equal to <strong>the</strong> logical<br />
definition,<br />
I ∧A ∧O R, cited earlier in chapter three.<br />
There is more to say about <strong>the</strong> structure <strong>of</strong> model fragments. <strong>Model</strong> fragments describe parts <strong>of</strong><br />
systems and can be ei<strong>the</strong>r structural or behavioural descriptions (process fragments). A structural or<br />
process description may in itself be a part <strong>of</strong> a more complex system part. A flow process for example<br />
is usually part <strong>of</strong> a system component where descriptions <strong>of</strong> a source and destination are required. By<br />
describing <strong>the</strong> distinct parts <strong>of</strong> a system in a generic fashion, <strong>the</strong> descriptions become reusable for<br />
o<strong>the</strong>r system model descriptions as well. GARP, for instance, just searches for model fragments by<br />
comparing a given description <strong>of</strong> some elements to <strong>the</strong> conditions <strong>of</strong> existing model fragments. It than<br />
automatically completes such a system model description with model fragments that have <strong>the</strong> same<br />
elements named in <strong>the</strong> condition part. The same model fragments can be instantiated on <strong>the</strong> basis <strong>of</strong><br />
different system model descriptions, and this advantage is maximised when <strong>the</strong> model fragments<br />
contain generic information. <strong>Model</strong> fragments can be made more generic by using a hierarchy <strong>of</strong><br />
model fragments. Very basic principles can than be extracted from more complex descriptions and a<br />
set <strong>of</strong> simple descriptions may be combined to form a complex structure. In ei<strong>the</strong>r way, <strong>the</strong> use <strong>of</strong><br />
generic knowledge bans out redundant descriptions <strong>of</strong> similar principles, and results in generally<br />
applicable descriptions. System model descriptions represent <strong>the</strong> state <strong>of</strong> a system at a particular<br />
moment in time, and are <strong>the</strong> effect <strong>of</strong> <strong>the</strong> actions described in a model fragment. A scenario, or input<br />
system, is a specific type <strong>of</strong> system model description. It is typically an incomplete description <strong>of</strong> <strong>the</strong><br />
real world system, from which <strong>the</strong> simulator tries to generate a full state description with all relevant<br />
system elements, partial models, etc that must be instantiated according to <strong>the</strong> conditions in <strong>the</strong><br />
scenario. The actual change from one system model description (a qualitative state) to <strong>the</strong> next is<br />
determined by <strong>the</strong> library <strong>of</strong> transformation rules. It is sufficient to say for <strong>the</strong> moment that <strong>the</strong><br />
transformation rules resolve all <strong>the</strong> parameter relations and determine which object and parameters<br />
will remain throughout <strong>the</strong> next state and which not (termination rules), that some terminations must<br />
be ordered or merged (precedence rules) and <strong>the</strong> things that have not been part <strong>of</strong> any termination must<br />
be unchanged in <strong>the</strong> next state (continuity rules).<br />
20
5. <strong>Model</strong> construction<br />
A few issues will be raised on <strong>the</strong> design <strong>of</strong> a model, and design methods in particular. Although some<br />
research has been conducted on which particular knowledge representation frameworks can fit a<br />
particular model design problem (Schut & Bredeweg 1996), much less is written down on subjects<br />
involving how <strong>the</strong> formal representation language and domain knowledge syn<strong>the</strong>sise into a coherent<br />
model, <strong>the</strong> process <strong>of</strong> designing. Somewhat typically, most resources in this respect can be found in<br />
<strong>the</strong> field <strong>of</strong> Computer Aided Design. In this area, a lot <strong>of</strong> research is done on developing formal and<br />
structural accounts for <strong>the</strong> design process, a prerequisite for automating <strong>the</strong> design process. The result<br />
is a bulk <strong>of</strong> literature that attempts to characterise methods used for design. In <strong>the</strong> following some<br />
issues will be discussed that seem useful in <strong>the</strong> current scope; to characterise and define <strong>the</strong> design<br />
process that is conducted to map <strong>the</strong> features <strong>of</strong> <strong>the</strong> water cycle to a qualitative simulation model.<br />
5.1 <strong>Model</strong> design process<br />
The design process is an exercise in mapping needs to function to structure, and in general <strong>the</strong> process<br />
deals with <strong>the</strong> satisfaction <strong>of</strong> constraints imposed by different actors, thus, it may be considered a<br />
problem solving process. A design space has as input some knowledge source and some problemsolving<br />
strategies. In <strong>the</strong> article on intelligent computer-aided design (D.C. Brown 1996), Brown lists<br />
several common <strong>of</strong> types <strong>of</strong> design. The most standard types <strong>of</strong> design are routine, innovative and<br />
creative design, where each successive type requires less preliminary knowledge in terms <strong>of</strong> <strong>the</strong> above<br />
named knowledge sources or problem solving strategy. O<strong>the</strong>r characterisations <strong>of</strong> <strong>the</strong>se design types<br />
focus on what is known about <strong>the</strong> variables and <strong>the</strong>ir values. From this point <strong>of</strong> view, an innovative<br />
design for example is done with fixed variables and changing values, see also (A. K. Goel, 1997).<br />
<strong>An</strong>o<strong>the</strong>r typology <strong>of</strong> <strong>the</strong> design process mentioned by Brown is conceptual versus parametric design.<br />
The former depends on abstract specifications <strong>of</strong> requirements, whereas <strong>the</strong> latter is more focused in<br />
terms <strong>of</strong> specification <strong>of</strong> parameters, values, and variables. A special form <strong>of</strong> <strong>the</strong> parametric design<br />
method is <strong>the</strong> propose and revise method, comparable to iterative design. In ano<strong>the</strong>r form <strong>of</strong> design,<br />
configuration, <strong>the</strong> design language is already pre-specified, and components <strong>of</strong> <strong>the</strong> language can be put<br />
toge<strong>the</strong>r to satisfy requirements.<br />
A special kind <strong>of</strong> creative design involves exploring <strong>the</strong> utilities <strong>of</strong> using analogies in proposing<br />
solutions to design problems in <strong>the</strong> conceptual phase <strong>of</strong> <strong>the</strong> design process (A.K. Goel, 1997).<br />
Following his definition, “..analogical design involves reminding and transfer <strong>of</strong> elements <strong>of</strong> a<br />
solution (S_old) for one design problem (P_old) to <strong>the</strong> solution (S_new) for ano<strong>the</strong>r design problem<br />
(P_new), where <strong>the</strong> selected design elements can be components, relations between components, or<br />
21
configurations <strong>of</strong> components and relations.” Fur<strong>the</strong>rmore, Goel poses four questions in considering<br />
analogical design; Why? What? How? and When?. The questions pertain to <strong>the</strong> tasks, content, methods<br />
and control <strong>of</strong> processing when using analogy in design. By asking why to use analogical design<br />
methods, it is possible to think <strong>of</strong> a number <strong>of</strong> design tasks that it may support, including proposing<br />
new solutions, decomposition, anticipating possible difficulties etc. What can be analogically designed<br />
refers to <strong>the</strong> content <strong>of</strong> <strong>the</strong> model, e.g. components or relations between components, and it may<br />
support design proposition or modification. <strong>An</strong> issue that is certainly related to this type <strong>of</strong> analogical<br />
design, which refers to domain or strategic knowledge is <strong>the</strong> problem <strong>of</strong> shared knowledge, more<br />
particular shared logic, ontology and/or computation pointed out by (Sowa, 2000). Certainly on <strong>the</strong><br />
ontological level “…different systems may use names for <strong>the</strong> same kinds <strong>of</strong> entities; even worse, <strong>the</strong>y<br />
may use <strong>the</strong> same names for different kinds.” (Sowa, 2000; p. 408).<br />
A common answer to how to perform analogical design is <strong>the</strong> use <strong>of</strong> case-based reasoning. In <strong>the</strong><br />
literature, distinction is made between transformational and derivational case-based reasoning, where<br />
<strong>the</strong> latter not only transfers and modifies S_old to S_new, but <strong>the</strong> trace <strong>of</strong> problem-space search that<br />
led to <strong>the</strong> solution S_old for P_old is included. The general flow <strong>of</strong> <strong>the</strong> tasks <strong>of</strong> <strong>the</strong> design process can<br />
be characterised as:<br />
Requirements Formulation Syn<strong>the</strong>sis <strong>An</strong>alysis Evaluation<br />
O<strong>the</strong>r perspectives see <strong>the</strong> design as incremental refinement, for example (Tong 1987). One <strong>of</strong> <strong>the</strong> core<br />
tasks in design, however, is <strong>the</strong> task <strong>of</strong> decomposition and especially decomposition guided by <strong>the</strong> use<br />
<strong>of</strong> hierarchies; see (Lee et al. 1992; Umeda & Tomiyama) for a detailed discussion on types <strong>of</strong><br />
hierarchies and decomposition. A whole range <strong>of</strong> decomposition methods can be used in <strong>the</strong> design<br />
process, ranging from structural decomposition to functional decomposition or behavioural<br />
decomposition (for example an influence diagram). In addition, <strong>the</strong> direction <strong>of</strong> processing with<br />
respect to <strong>the</strong> development <strong>of</strong> models is a issue <strong>of</strong> interest. Both bottom-up and top-down ways <strong>of</strong><br />
processing may be combined to resolve goals. <strong>An</strong> example is <strong>the</strong> hybrid-opportunistic technique<br />
mentioned by (Stefik, 1981). In terms <strong>of</strong> knowledge types <strong>the</strong> use <strong>of</strong> ontologies can provide<br />
clarification <strong>of</strong> <strong>the</strong> concepts and relations within a domain <strong>of</strong> application. Besides functioning as<br />
domain-specific catalog <strong>of</strong> concepts, ano<strong>the</strong>r virtue <strong>of</strong> ontologies concerns <strong>the</strong> sharing and reuse <strong>of</strong><br />
knowledge between various actors. The latter are usually referred to as task ontologies. The techniques<br />
associated with qualitative modelling may be considered as task ontology in this respect.<br />
22
5.2 Classification<br />
A rigid classification, or ontology, <strong>of</strong> <strong>the</strong> things in <strong>the</strong> world has not been agreed upon as <strong>of</strong> yet. Most<br />
<strong>of</strong> <strong>the</strong> philosophers from our time and from <strong>the</strong> past have made propositions about how to classify <strong>the</strong><br />
things around us. So it is not very surprisingly that <strong>the</strong> qualitative or naive physics enterprise has some<br />
holistic features as well. This means that entities, structures, relations or conceptual clusters in general<br />
cannot be defined indiscriminately from o<strong>the</strong>rs. A related and sometimes even more distressing issue<br />
concerns <strong>the</strong> characteristics <strong>of</strong> entities and <strong>the</strong>ir relation with respect to space and time, which<br />
sometimes need to be defined. A discussion <strong>of</strong> shortcomings <strong>of</strong> <strong>the</strong> (computational) application <strong>of</strong><br />
common sense <strong>the</strong>ory in knowledge-intensive systems is given in (Smith & Casati, 1994). The<br />
attempts to produce realistic <strong>the</strong>ories <strong>of</strong> <strong>the</strong> ‘naive’ physical world using formal <strong>the</strong>ories and logical<br />
inferences seem to be in some contrast with <strong>the</strong>ories <strong>of</strong> <strong>the</strong> common sense world as defined by for<br />
example Gestalt psychologists. There exists a risk that <strong>the</strong> purpose <strong>of</strong> efficient representations will<br />
lead to <strong>the</strong> neglect <strong>of</strong> <strong>the</strong> contributions <strong>of</strong> o<strong>the</strong>r disciplines to <strong>the</strong> <strong>the</strong>ories <strong>of</strong> common sense.<br />
From <strong>the</strong> current point <strong>of</strong> view, an approach to classifying phenomena is needed to distinguish <strong>the</strong><br />
concepts we want to incorporate in our model. Thus when building a model for a particular purpose,<br />
decisions have to be made on which ontology is used. A more or less common list <strong>of</strong> (relatively<br />
discriminable) clusters in naive physics is <strong>the</strong> following (see for example Hayes 1979, pp. 187-97):<br />
• Objects, Natural Kinds<br />
• Events, Processes and Causality<br />
• Stuffs, States <strong>of</strong> Matter, Qualities<br />
• Surfaces, Limits, Boundaries<br />
The list specified above gives handles to start decomposing <strong>the</strong> concepts we engage in a particular<br />
domain in <strong>the</strong> world. But we also need criteria or definitions <strong>of</strong> <strong>the</strong> clusters in order to assign a<br />
particular concept to a cluster. This task is commonly referred to as ontological engineering (Sowa,<br />
2000). As for <strong>the</strong> formalisation <strong>of</strong> naive physics, <strong>the</strong>re exist ontologies to classify concepts such as <strong>the</strong><br />
ones mentioned above, for example <strong>the</strong> ontology for liquids developed by Patrick Hayes (Hayes,<br />
1978) or <strong>the</strong> qualitative process <strong>the</strong>ory by Kenneth Forbus (Forbus, 1983). In modelling <strong>the</strong> water<br />
cycle we will use aspects <strong>of</strong> <strong>the</strong> formal ontologies previously mentioned in order to classify <strong>the</strong><br />
domain.<br />
23
5.3 Organisation <strong>of</strong> models and ontology<br />
So far <strong>the</strong> focus has been on <strong>the</strong>ories about how to represent different types <strong>of</strong> knowledge that people<br />
use in reasoning and problem solving. Qualitative reasoning methods in general and <strong>the</strong> compositional<br />
modelling approach in particular well suited to represent phenomena that are associated with physical<br />
processes. In terms <strong>of</strong> <strong>the</strong> formal encoding <strong>of</strong> qualitative knowledge, <strong>the</strong> possibility <strong>of</strong> reusing partial<br />
models has been discussed and in terms <strong>of</strong> model design <strong>the</strong> concept <strong>of</strong> analogical design and<br />
ontologies have been given some attention in <strong>the</strong> last paragraph. In this paragraph we will turn to <strong>the</strong><br />
content <strong>of</strong> <strong>the</strong> model, by looking at how knowledge is classified (ontology) and which organisation <strong>of</strong><br />
knowledge fits <strong>the</strong> type <strong>of</strong> model (i.e. didactic) that is developed.<br />
Usually, multiple perspectives apply to a certain knowledge domain. Different perspectives for<br />
example may emphasise various levels <strong>of</strong> analysis with respect to granularity <strong>of</strong> a domain. Being able<br />
to reason about a domain from different perspectives typically should improve <strong>the</strong> understanding <strong>of</strong> a<br />
domain.<br />
<strong>An</strong>o<strong>the</strong>r issue, particularly relevant for didactic systems is <strong>the</strong> dimensions along which knowledge is<br />
(re)presented. Such an approach is outlined by (Salles & Bredeweg, 2001) by <strong>the</strong> idea <strong>of</strong> constructing<br />
progressive learning routes in <strong>the</strong> domain <strong>of</strong> <strong>the</strong> Cerrado vegetation. They propose <strong>the</strong> implementation<br />
<strong>of</strong> learning routes that evolve from static to dynamic and from less complex to more complex.<br />
Therefore <strong>the</strong> simulation model <strong>of</strong> <strong>the</strong> domain is subdivided in smaller units that can be termed model<br />
dimensions. The model dimensions that are defined are zero-order (static knowledge), first order<br />
(behaviour), and second order (relative behaviour). Mapping <strong>the</strong> subject matter to student’s mental<br />
model clearly is an issue that has to be incorporated in <strong>the</strong> design <strong>of</strong> <strong>the</strong> domain knowledge simulator.<br />
Implementing <strong>the</strong>se views contributes to keeping <strong>the</strong> interaction between <strong>the</strong> student’s mental model<br />
and <strong>the</strong> simulation on <strong>the</strong> same perspective and granularity. The approach is an integration <strong>of</strong> various<br />
perspectives on organising <strong>the</strong> subject matter model, see for example causal model progression by<br />
(White & Frederiksen) and <strong>the</strong> didactic goal generator by (Winkels & Breuker). It is argued that <strong>the</strong><br />
resulting clusters <strong>of</strong> models will support knowledge communication with qualitative models in<br />
ecological domains to a better extent.<br />
In practice, an effort has to be made in terms <strong>of</strong> ontological engineering with respect to classifying <strong>the</strong><br />
knowledge in <strong>the</strong> domain if we want to be consistent in issuing knowledge to <strong>the</strong> student along various<br />
dimensions. It means that it is not only necessary to define <strong>the</strong> categories and dimensions <strong>of</strong><br />
knowledge, but also a consensus about <strong>the</strong> method <strong>of</strong> actually classifying <strong>the</strong> knowledge under certain<br />
headings, a typology <strong>of</strong> <strong>the</strong> knowledge. Related to this observation is <strong>the</strong> article by Mizoguchi and<br />
Bourdeau (2000) in which <strong>the</strong>y describe <strong>the</strong> use <strong>of</strong> ontological engineering to overcome common AI-<br />
ED problems. They observe <strong>the</strong> lack <strong>of</strong> consensus about <strong>the</strong> conceptualisation <strong>of</strong> knowledge in various<br />
knowledge intensive systems, and <strong>the</strong>y argue for explicit representation <strong>of</strong> <strong>the</strong> conceptualisation on<br />
24
which <strong>the</strong> system is based on, and to a certain extent even a standardisation. The hypo<strong>the</strong>sis is that it<br />
will support <strong>the</strong> adaptability and reuse <strong>of</strong> various components <strong>of</strong> knowledge intensive systems. They<br />
argue for abandoning a mere heuristic-driven knowledge engineering toward increasingly model-based<br />
approaches. ”The power lies in <strong>the</strong> knowledge” proposed by Feigenbaum is a principle that suits <strong>the</strong><br />
vision. This new approach to knowledge modelling has it’s origins in projects as KADS and<br />
PROTEGE, and it introduces <strong>the</strong> notions <strong>of</strong> ontologies, being divided into a task ontology, which<br />
specifies <strong>the</strong> problem solving architecture <strong>of</strong> knowledge-based systems, and domain ontology which<br />
specifies <strong>the</strong> domain knowledge. The idea <strong>of</strong> this knowledge sharing between people and computers at<br />
<strong>the</strong> conceptual level is not new. The approach <strong>of</strong> Brown & de Kleer (1984), a framework for a<br />
qualitative physics, is an approach that puts effort into defining conceptualisations that are used for<br />
representing <strong>the</strong> world in a qualitative way.<br />
5.4 Qualitative modelling <strong>of</strong> a domain<br />
"Solving a design problem efficiently requires an adequate representation (…). One <strong>of</strong> <strong>the</strong> major tasks<br />
(…) is to identify all <strong>the</strong> pieces <strong>of</strong> knowledge used by <strong>the</strong> system and map <strong>the</strong>m onto <strong>the</strong> most<br />
convenient representation provided by AI." (D.C. Brown & D.L. Grecu, 1996).<br />
According to Schut & Bredeweg (1996), in <strong>the</strong>ir article on approaches to qualitative model<br />
construction, we can distinguish between <strong>the</strong> real world (<strong>the</strong> referent), and <strong>the</strong> model <strong>of</strong> <strong>the</strong> real world.<br />
A model <strong>of</strong> a referent can be made to represent certain chosen features <strong>of</strong> <strong>the</strong> referent, and <strong>the</strong> goal <strong>of</strong><br />
representing this information is that it can be retrieved more efficiently. Typical to model construction<br />
thus is <strong>the</strong> task <strong>of</strong> selection <strong>of</strong> relevant features to be incorporated in <strong>the</strong> model. Specialising on<br />
qualitative models, <strong>the</strong>ir aim is to compare two approaches to qualitative model building: <strong>the</strong> model<br />
composition approach, and <strong>the</strong> model induction approach, and consider <strong>the</strong> comparison a general<br />
frame <strong>of</strong> reference for comparing approaches to model building in a qualitative way.<br />
In modelling a domain, <strong>the</strong> main focus lies in <strong>the</strong> relation between <strong>the</strong> referent and <strong>the</strong> model, which is<br />
a representation relation. This means that for some distinguished aspects at <strong>the</strong> least, <strong>the</strong> referent and<br />
<strong>the</strong> model must correspond. The things from <strong>the</strong> referent that will be represented in <strong>the</strong> model are thus<br />
dependent on <strong>the</strong> representation primitives available in <strong>the</strong> reasoning formalism (see chapter 4), and<br />
also on <strong>the</strong> purpose <strong>of</strong> <strong>the</strong> model. The way in which representational primitives shape <strong>the</strong><br />
representation relation is important to <strong>the</strong> conceptual closure <strong>of</strong> <strong>the</strong> model (Schut & Bredeweg 1996).<br />
The purpose <strong>of</strong> <strong>the</strong> model requires a typical way <strong>of</strong> conceptualising <strong>the</strong> referent, which means that <strong>the</strong><br />
purpose <strong>of</strong> <strong>the</strong> model specifies <strong>the</strong> perspective taken on <strong>the</strong> referent.<br />
In this research, <strong>the</strong> goal is <strong>the</strong> design <strong>of</strong> a qualitative simulation model for educational purposes.<br />
More specifically, <strong>the</strong> educational part in this research is directed to facilitating progressive learning<br />
routes through a qualitative simulation <strong>of</strong> <strong>the</strong> water cycle. In technical terms, <strong>the</strong> educational<br />
25
perspective requires <strong>the</strong> domain knowledge to be characterised by clusters <strong>of</strong> models, which vary with<br />
respect to certain dimensions such as granularity and complexity. The input scenarios to <strong>the</strong> simulator<br />
can be structured according to <strong>the</strong>se principles, so that learners are confronted with models <strong>of</strong><br />
increasing elaboration.<br />
<strong>An</strong> important issue concerns <strong>the</strong> multiple views that may coexist in one model. <strong>Multiple</strong> views allow<br />
<strong>the</strong> same phenomena to be analysed on different levels <strong>of</strong> granularity or complexity. In more advanced<br />
compositional modelling <strong>the</strong> mapping from <strong>the</strong> scenario to <strong>the</strong> library <strong>of</strong> model fragments can be oneto-many.<br />
This has <strong>the</strong> effect that <strong>the</strong> generated model can be adjusted to <strong>the</strong> desired level <strong>of</strong> detail, i.e.<br />
<strong>the</strong> scope <strong>of</strong> <strong>the</strong> model may differ. These different levels <strong>of</strong> detail or completeness are called model<br />
dimensions. For this, multiple incompatible views on a phenomenon must exist side by side in <strong>the</strong><br />
library, and <strong>the</strong> choice whe<strong>the</strong>r one or ano<strong>the</strong>r model is selected depends on <strong>the</strong> information that is<br />
described by <strong>the</strong> goal for which <strong>the</strong> model is used. This information can be captured in formulating<br />
explicit model assumptions under which <strong>the</strong> model is applicable. The model assumptions involved in a<br />
particular simulation are characterised by taking <strong>the</strong> particular perspective on a system. The idea is<br />
depicted in figure 5.1.<br />
Figure 5.1. A central role for perspective.<br />
The representation relation between <strong>the</strong> model and <strong>the</strong> referent can be adjusted by comparing <strong>the</strong><br />
output <strong>of</strong> a particular simulation to <strong>the</strong> behaviour that is found in <strong>the</strong> referent. By using <strong>the</strong> procedure<br />
26
<strong>of</strong> hypo<strong>the</strong>se and test, <strong>the</strong> model can be tuned more precisely to <strong>the</strong> phenomena it is supposed to<br />
represent.<br />
<strong>Model</strong>ling a referent means depicting a real world system in terms <strong>of</strong> a (chosen) formalism. The<br />
modelling framework that is <strong>the</strong> result <strong>of</strong> a particular conceptualisation <strong>of</strong> <strong>the</strong> referent consists <strong>of</strong> a<br />
conceptual framework and an inference procedure. The conceptual framework must be materialised in<br />
<strong>the</strong> representation and reasoning techniques <strong>of</strong> qualitative reasoning.<br />
Figure 5.2. Tasks in model construction.<br />
Figure 5.2 specifies some general tasks that are involved with translating conceptual ideas into <strong>the</strong><br />
qualitative language before implementing and simulating <strong>the</strong> model. The behavioural description<br />
focuses on <strong>the</strong> parts <strong>of</strong> <strong>the</strong> referent that are relevant to <strong>the</strong> causal chain <strong>of</strong> events. In order to give <strong>the</strong>se<br />
parts <strong>of</strong> <strong>the</strong> system a qualitative representation, a general inventory <strong>of</strong> processes and objects has to be<br />
made. Variables and <strong>the</strong> relations between certain variables are specified qualitatively as well,<br />
including ideas on what quantity spaces should be assigned. The implementation concerns <strong>the</strong><br />
production <strong>of</strong> a library containing model fragments, an isa-hierarchy containing entities and a set <strong>of</strong><br />
input systems that form <strong>the</strong> different scenario’s in <strong>the</strong> simulation. The task after that is analysing if <strong>the</strong><br />
output <strong>of</strong> <strong>the</strong> simulation matches <strong>the</strong> referent for <strong>the</strong> intended relevant aspects<br />
27
6. The <strong>Water</strong> <strong>Cycle</strong><br />
With <strong>the</strong> goal <strong>of</strong> developing a qualitative model <strong>of</strong> <strong>the</strong> water cycle, capable <strong>of</strong> reasoning about<br />
causality and physical processes, <strong>the</strong> need for a mapping between <strong>the</strong> naive mental model and <strong>the</strong><br />
computational model <strong>of</strong> a domain was addressed in <strong>the</strong> previous paragraphs, and <strong>the</strong> issues <strong>of</strong><br />
knowledge engineering that are related to it. Before being able to <strong>the</strong>se issues onto <strong>the</strong> particular<br />
domain <strong>of</strong> hydrology, an exploration <strong>of</strong> <strong>the</strong> fields <strong>of</strong> hydrology and meteorology must precede.<br />
Although <strong>the</strong> field <strong>of</strong> meteorology has been used for a number <strong>of</strong> applications in ITS, in <strong>the</strong> more<br />
recent work on qualitative reasoning only (Iwasaki, 1996) developed a system for answering<br />
prediction questions <strong>of</strong> rainfall in Japan, and it does not attempt to give a causal account <strong>of</strong> <strong>the</strong> water<br />
cycle in a more general way. <strong>An</strong> older example is <strong>the</strong> METEOROLOGY tutor, a system that made use<br />
<strong>of</strong> simulation based reasoning, where causality is implemented in sequences <strong>of</strong> events which are<br />
simulated by transition between states. On <strong>the</strong> o<strong>the</strong>r hand, a reasonable amount <strong>of</strong> research on<br />
qualitative representations has been performed in <strong>the</strong>rmodynamical domains, see for example (Collins<br />
& Forbus 1989, Pisan, 1996). From <strong>the</strong> examples in <strong>the</strong> literature one can readily see that qualitative<br />
representation fits <strong>the</strong> type <strong>of</strong> domains that involve heat exchanges, substance flows phase changes,<br />
etc. well. The importance <strong>of</strong> processes and causality are apparent, and <strong>the</strong>se can be described in an<br />
explicit manner, which is applicable to <strong>the</strong> <strong>the</strong>ories <strong>of</strong> qualitative physics in general and qualitative<br />
process <strong>the</strong>ory in particular. After describing <strong>the</strong> functioning <strong>of</strong> <strong>the</strong> water cycle it will be translated<br />
into a complete qualitative account.<br />
6.1 Domain description: Introduction to hydrology and meteorology.<br />
This discussion is meant to give an overview <strong>of</strong> <strong>the</strong> main processes and concepts that govern <strong>the</strong> water<br />
cycle. As such it deals with substances in <strong>the</strong> air as well as water on <strong>the</strong> surface. It is constantly tried<br />
to give a causal description <strong>of</strong> <strong>the</strong> processes, for <strong>the</strong> simple reason that a good causal account is a main<br />
ingredient in developing a knowledge-intensive simulation model <strong>of</strong> <strong>the</strong> water cycle. It must be seen to<br />
that all <strong>the</strong> factors that have a direct influence on <strong>the</strong> causal chain <strong>of</strong> events will be mentioned, ins<strong>of</strong>ar<br />
<strong>the</strong>y are relevant to <strong>the</strong> learning goals. In <strong>the</strong> following stadium, it will be this description that will<br />
serve as reference from which <strong>the</strong> choices in knowledge representation and modelling assumptions<br />
will be argued, besides <strong>the</strong> framework <strong>of</strong> rules and constraints that will be posed on it by <strong>the</strong> <strong>the</strong>ories<br />
<strong>of</strong> qualitative physics and automated modelling.<br />
The water cycle is a dynamic process that governs <strong>the</strong> circulation <strong>of</strong> water on earth. The processes that<br />
occur in <strong>the</strong> atmosphere are part <strong>of</strong> <strong>the</strong> meteorological science, which is also occupied with describing<br />
<strong>the</strong> wea<strong>the</strong>r on earth. The three main processes that occur in <strong>the</strong> atmosphere relevant to <strong>the</strong> water cycle<br />
28
are evaporation, condensation and precipitation. A fourth process, transport <strong>of</strong> water in <strong>the</strong> air, plays a<br />
big role in <strong>the</strong> distribution <strong>of</strong> <strong>the</strong> water throughout <strong>the</strong> atmosphere. In <strong>the</strong> atmosphere, water can be<br />
present in <strong>the</strong> form <strong>of</strong> gas (vapour), liquid, and ice. Just like <strong>the</strong> air itself, water has a certain pressure<br />
that acts on a certain volume <strong>of</strong> air, which is indicated by vapour pressure. Under some circumstances<br />
<strong>the</strong> vapour pressure may reach a limit value, called <strong>the</strong> saturation vapour pressure. The saturation<br />
vapour pressure increases with higher temperatures, because warm air expands and can hold more<br />
vapour. It is also true that unsaturated air can become saturated when it cools down.<br />
Whenever <strong>the</strong> saturation vapour pressure is reached, any fur<strong>the</strong>r evaporation <strong>of</strong> liquid water in that<br />
part <strong>of</strong> <strong>the</strong> air will cause condensation <strong>of</strong> <strong>the</strong> vapour. <strong>An</strong>o<strong>the</strong>r measure to discover <strong>the</strong> amount <strong>of</strong> moist<br />
that <strong>the</strong> air can hold is called <strong>the</strong> relative humidity, which is <strong>the</strong> ratio <strong>of</strong> <strong>the</strong> actual vapour pressure to<br />
<strong>the</strong> saturation vapour pressure. thus it represents <strong>the</strong> amount <strong>of</strong> moisture in a space compared to <strong>the</strong><br />
amount it could hold if it were saturated. The temperature at which vapour starts to condense is called<br />
<strong>the</strong> dew point, and <strong>the</strong> combination <strong>of</strong> temperature (T, in degrees Celsius) and <strong>the</strong> dew point serves as<br />
an indication <strong>of</strong> how far <strong>the</strong> temperature <strong>of</strong> <strong>the</strong> air is from <strong>the</strong> point where it will be saturated.<br />
Cloud formation is <strong>the</strong> result <strong>of</strong> water vapour being condensed. The transformation from vapour to<br />
water droplets and vice versa requires an energy-transfer. Energy is needed to evaporate water (like in<br />
boiling) but inversely, it is also true that <strong>the</strong> condensation <strong>of</strong> vapour releases energy to <strong>the</strong><br />
environment. So if for example a droplet falls through an unsaturated layer <strong>of</strong> air, <strong>the</strong> droplet can<br />
vaporise and <strong>the</strong>reby cool down <strong>the</strong> air layer. This also means that air can become saturated at<br />
temperatures above <strong>the</strong> dew point, as a consequence <strong>of</strong> an increase in water supply. The most common<br />
cause <strong>of</strong> air becoming saturated is <strong>the</strong> cooling <strong>of</strong> air. For water droplets to form, i.e. condensation, <strong>the</strong><br />
condition is not only that <strong>the</strong> air is saturated, but also <strong>the</strong> presence <strong>of</strong> so-called condensation nuclei,<br />
which are very tiny particles <strong>of</strong> salt or acid solutions in <strong>the</strong> atmosphere. The water molecules will<br />
attach to <strong>the</strong>se nuclei. <strong>An</strong> important thing is also that <strong>the</strong> normal cloud droplets are not big enough to<br />
reach earth's surface before being evaporated again. The droplets that we call rain are much bigger<br />
than <strong>the</strong> cloud droplets, <strong>the</strong> main reason for this is that bigger droplets catch smaller droplets and grow<br />
because <strong>of</strong> that. Through continuous movement <strong>of</strong> <strong>the</strong> water in <strong>the</strong> clouds some droplets will grow<br />
until <strong>the</strong>y are so big that <strong>the</strong> air cannot hold <strong>the</strong>m anymore.<br />
Condensation is <strong>the</strong> direct cause <strong>of</strong> all forms <strong>of</strong> precipitation. It occurs under various conditions that in<br />
one way or ano<strong>the</strong>r are associated with a change in one <strong>of</strong> <strong>the</strong> linked parameters <strong>of</strong> air, volume,<br />
temperature, pressure and humidity. This brings us to <strong>the</strong> movement <strong>of</strong> air in <strong>the</strong> atmosphere. The<br />
vertical movement <strong>of</strong> air can be explained with <strong>the</strong> specific gravity <strong>of</strong> various volumes <strong>of</strong> air. Air<br />
volumes with lower specific gravity will rise and air volumes with higher specific gravity will sink.<br />
The difference in specific weight is solely caused by difference in temperature. <strong>An</strong> air volume with<br />
higher temperature has a lower specific gravity and vice versa. Extra complications with vertical<br />
29
movements are <strong>the</strong> difference in temperature at various height in <strong>the</strong> atmosphere and <strong>the</strong> change <strong>of</strong><br />
temperature within <strong>the</strong> vertically moving volume <strong>of</strong> air. The latter is <strong>the</strong> consequence <strong>of</strong> <strong>the</strong> fact that a<br />
rising volume <strong>of</strong> air will be surrounded by lower pressure and a sinking air volume will have higher<br />
pressure in <strong>the</strong> surrounding. In <strong>the</strong> first case <strong>the</strong> volume <strong>of</strong> air will expand and cool, in <strong>the</strong> second case<br />
it will be compressed and <strong>the</strong>refore heated. The change in temperature that occurs within <strong>the</strong> vertical<br />
moving air is <strong>the</strong>refore not caused by external heat supplies and called adiabatic movement. The<br />
relative difference between <strong>the</strong> temperature <strong>of</strong> a volume <strong>of</strong> air and <strong>the</strong> temperature <strong>of</strong> its surroundings<br />
is called <strong>the</strong> vertical temperature gradient. The ratio <strong>of</strong> this gradient will ei<strong>the</strong>r repress or encourage<br />
vertical movements. In <strong>the</strong> fist case <strong>the</strong> air is said to be stable, and in <strong>the</strong> second case air is said to be<br />
unstable. <strong>the</strong> occurrence <strong>of</strong> condensation within <strong>the</strong>se vertical movements means that <strong>the</strong> air is<br />
saturated. if saturated air is forced to rise, it will cool and <strong>the</strong> water would become supersaturated.<br />
Instead, <strong>the</strong> water will condense to <strong>the</strong> extent that <strong>the</strong> saturation vapour pressure is maintained at <strong>the</strong><br />
corresponding temperature. This process releases heat to <strong>the</strong> air volume which positively affect <strong>the</strong><br />
elevation. On <strong>the</strong> o<strong>the</strong>r hand, if a condense-holding volume <strong>of</strong> air is forced downwards, it will become<br />
unsaturated and some <strong>of</strong> <strong>the</strong> condense turns into vapour again and <strong>the</strong> air volume cools down.<br />
In order for water to evaporate, a heat source must be present. This can be <strong>the</strong> sun, an overlying air<br />
layer that is warmer, <strong>the</strong> ground and water itself. Note that <strong>the</strong> sun is an external influence to <strong>the</strong> cycle<br />
in this respect. Evaporation is <strong>the</strong> process by which a liquid or a solid is changed to a gas, and energy<br />
has to be added to accomplish <strong>the</strong> change <strong>of</strong> state. Ultimately <strong>the</strong> source <strong>of</strong> this energy is current or<br />
recent solar energy impinging on water, <strong>the</strong> soil-water complex, and <strong>the</strong> plant-life-water complex.<br />
Evaporation <strong>of</strong> water from a given surface is greatest when <strong>the</strong> air is warm and dry, and smallest when<br />
<strong>the</strong> air is cold and moist, because in warm conditions, <strong>the</strong> saturation deficit is large and conversely in<br />
cold conditions it is small. There is thus an underlying relation between <strong>the</strong> rate <strong>of</strong> evaporation and <strong>the</strong><br />
saturation deficit. Since <strong>the</strong> process <strong>of</strong> evaporation involves <strong>the</strong> net movement <strong>of</strong> water vapour<br />
molecules from an evaporating surface into <strong>the</strong> overlying air, it will eventually lead to <strong>the</strong> saturation <strong>of</strong><br />
<strong>the</strong> lowest air layers <strong>of</strong> overlaying air and <strong>the</strong> termination <strong>of</strong> evaporation. This termination <strong>of</strong><br />
evaporation mostly does not occur, because <strong>of</strong> convective movements, which create wind, and<br />
<strong>the</strong>refore reducing <strong>the</strong> water vapour content in <strong>the</strong> lowest layers, so that evaporation will continue.<br />
<strong>Water</strong> can evaporate from <strong>the</strong> oceans, lakes, rivers or from <strong>the</strong> leafs <strong>of</strong> plants. Toge<strong>the</strong>r <strong>the</strong>se<br />
processes are called evapotranspiration. Once in vaporised form <strong>the</strong> water will behave in accordance<br />
with <strong>the</strong> meteorological laws.<br />
The second major part <strong>of</strong> <strong>the</strong> water cycle concerns <strong>the</strong> processes that occur on <strong>the</strong> earth surface. Moist<br />
<strong>of</strong> precipitation that reaches <strong>the</strong> ground surface is usually absorbed by <strong>the</strong> surface layers <strong>of</strong> <strong>the</strong> soil.<br />
The remainder, once any depression storage has been filled, will flow over <strong>the</strong> surface as overland<br />
flow, reaching <strong>the</strong> stream channels quickly. The water that infiltrates into <strong>the</strong> soil may subsequently be<br />
30
evaporated or it may move under gravity and percolate downwards to <strong>the</strong> groundwater zone or else<br />
flow laterally close to <strong>the</strong> surface as through flow. The ability <strong>of</strong> <strong>the</strong> soil to absorb and retain moisture,<br />
<strong>the</strong> permeability, is crucial. Precipitation enters <strong>the</strong> soil at <strong>the</strong> ground surface and moves downward to<br />
<strong>the</strong> water table, which marks <strong>the</strong> upper surface <strong>of</strong> <strong>the</strong> zone <strong>of</strong> saturation. Immediately above <strong>the</strong> water<br />
table is <strong>the</strong> capillary fringe in which almost all <strong>the</strong> pores are full <strong>of</strong> water. The zone <strong>of</strong> aeration is a<br />
zone <strong>of</strong> transition in which water is absorbed, held or transmitted ei<strong>the</strong>r downwards toward <strong>the</strong> water<br />
table or upwards towards <strong>the</strong> soil surface, from where it can be evaporated.<br />
At a general level, <strong>the</strong> relation between run<strong>of</strong>f and precipitation can be regarded in terms <strong>of</strong> <strong>the</strong><br />
continuous circulation <strong>of</strong> water through <strong>the</strong> water cycle. Specifically, we can recognise that in natural<br />
conditions each river and stream receives water only from its own drainage basin or catchment area.<br />
Each catchment area can <strong>the</strong>refore be regarded as a system receiving inputs <strong>of</strong> precipitation and<br />
transforming <strong>the</strong>se into outputs <strong>of</strong> evaporation and run<strong>of</strong>f. Precipitation may arrive in <strong>the</strong> stream flows<br />
through ei<strong>the</strong>r, direct precipitation onto <strong>the</strong> water surface, through overland flow, through shallow<br />
subsurface flow (through flow),or through deep subsurface flow (groundwater flow).<br />
The process <strong>of</strong> infiltration concerns <strong>the</strong> water from precipitation that will be infiltrated in <strong>the</strong> soil.<br />
Effective rainfall is all <strong>the</strong> water that is left after infiltration and o<strong>the</strong>r losses as interception and<br />
evaporation. For water to infiltrate <strong>the</strong> soil must be permeable and not already saturated. if soil is<br />
saturated, a run<strong>of</strong>f over land is created.<br />
6.2 Conceptual model<br />
Having described <strong>the</strong> core features <strong>of</strong> interest within <strong>the</strong> hydrological domain, we turn to <strong>the</strong><br />
conceptual design <strong>of</strong> <strong>the</strong> simulation model. The basic idea is to form an abstract definition <strong>of</strong> <strong>the</strong><br />
objects and processes including <strong>the</strong>ir relations, which form <strong>the</strong> essentials <strong>of</strong> <strong>the</strong> domain. The water<br />
cycle as a system inhibits features <strong>of</strong> <strong>the</strong>rmodynamic systems including heat flows, phase changes,<br />
energy for example. From an educational point <strong>of</strong> view, <strong>the</strong> emphasis with introductions on<br />
meteorology usually lies on a set <strong>of</strong> core processes that govern <strong>the</strong> water cycle. The learner has to<br />
master (causal) models <strong>of</strong> evaporation, wind, precipitation and transport <strong>of</strong> air. In addition, <strong>the</strong> storage<br />
and flows <strong>of</strong> water on land are an integral part <strong>of</strong> <strong>the</strong> curriculum. These processes, depicted in fig 6.1,<br />
will form <strong>the</strong> central part <strong>of</strong> attention in <strong>the</strong> simulation model.<br />
31
Figure 6.1. Standard picture <strong>of</strong> <strong>the</strong> water cycle.<br />
The behaviour to be modelled is briefly <strong>the</strong> following: In order to get water from <strong>the</strong> sea to <strong>the</strong> air<br />
above, we need a heat source, in this case <strong>the</strong> sun, to evaporate <strong>the</strong> water. Once in gas form,<br />
continuous heating will cause <strong>the</strong> air with vapour to rise. The rise <strong>of</strong> air also causes cooling which in<br />
its turn causes <strong>the</strong> water to condense. Condensed water may be transported to parts <strong>of</strong> <strong>the</strong> air with<br />
lower pressure. Because transportation involves fur<strong>the</strong>r cooling, <strong>the</strong> water holding ability <strong>of</strong> <strong>the</strong> air<br />
will decrease. In <strong>the</strong> case <strong>of</strong> no water holding ability <strong>the</strong> precipitation process becomes active which<br />
will cause rain (precipitation). Once precipitation reaches <strong>the</strong> surface, it will be caught in storages on<br />
<strong>the</strong> surface, ei<strong>the</strong>r puddles or rivers. It can also fall directly back into <strong>the</strong> sea but that is not modelled at<br />
this time. Figure 6.2 forms <strong>the</strong> starting point <strong>of</strong> <strong>the</strong> conceptualisation, giving a global schematic idea<br />
<strong>of</strong> <strong>the</strong> water cycle.<br />
Figure 6.2. schematical representation <strong>of</strong> a water cycle.<br />
32
The schematic representation <strong>of</strong> <strong>the</strong> water cycle can be mapped onto qualitative representations. There<br />
exists previous work on qualitative modelling <strong>of</strong> <strong>the</strong>rmodynamic processes (Collins & Forbus, 1997;<br />
Skorstadt & Forbus, 1998), and contained stuffs in general (Hayes, 1979), in which certain analogies<br />
exist with <strong>the</strong> current research. The main difficulty here lies in putting <strong>the</strong> individual processes<br />
toge<strong>the</strong>r in a model that also inhibits more global domain knowledge <strong>of</strong> context and <strong>the</strong>refore also<br />
global constraints. This means that all processes should function given <strong>the</strong> same global set <strong>of</strong><br />
constraints. The first concern is <strong>the</strong> mapping <strong>of</strong> <strong>the</strong> entities in <strong>the</strong> world to entities that are qualitatively<br />
interpretable.<br />
6.3 Decomposition and models <strong>of</strong> different complexity<br />
When modelling a domain qualitatively, decisions have to be made on which particular parts <strong>of</strong> <strong>the</strong><br />
system can be described in a more or less independent way, complying with <strong>the</strong> no-function-instructure<br />
principle. <strong>Model</strong>ling also means specifying what components will be considered, i.e. 'exist',<br />
in <strong>the</strong> model, and this makes up <strong>the</strong> representation relation between model and referent, see section<br />
5.3. This section concerns <strong>the</strong> decomposition <strong>of</strong> <strong>the</strong> domain <strong>of</strong> <strong>the</strong> water cycle, and formalising <strong>the</strong><br />
components in model fragments. The model composition is done in different levels <strong>of</strong> detail. The<br />
ultimate goal is to combine several models and inspect different levels <strong>of</strong> detail using assumptions<br />
applying to different levels <strong>of</strong> analysis. With an educational point <strong>of</strong> view, controlling <strong>the</strong> level <strong>of</strong><br />
detail <strong>of</strong> a model facilitates curriculum design by allowing model progression. <strong>Model</strong> progression can<br />
be achieved by organising <strong>the</strong> content <strong>of</strong> a library along certain dimensions that represent particular<br />
levels <strong>of</strong> refinement <strong>of</strong> one’s knowledge <strong>of</strong> a system component. Dimensions depend on explicit<br />
definition <strong>of</strong> so-called building blocks (knowledge chunks) <strong>of</strong> <strong>the</strong> system, which again correspond to<br />
various levels <strong>of</strong> understanding and skill. As a consequence <strong>the</strong> general purpose <strong>of</strong> <strong>the</strong> model can be<br />
controlled, e.g. is it meant to reason on <strong>the</strong> level <strong>of</strong> expert or novice? Simple models can be extended<br />
and altered to a certain extent to create bigger and more detailed models. In this case, two separate<br />
models <strong>of</strong> <strong>the</strong> water cycle have been made. The first model incorporates <strong>the</strong> idea <strong>of</strong> <strong>the</strong> core processes<br />
<strong>of</strong> <strong>the</strong> water cycle in a very abstract and simplified manner. For example, influences from <strong>the</strong><br />
atmosphere and height differences are not being modelled. The second model introduces more<br />
knowledge and reasons more precisely and detailed about certain processes.<br />
33
6.3.1 Homer<br />
Homer is a model building environment that allow users to construct qualitative models using visual<br />
representations <strong>of</strong> <strong>the</strong> entities and primitives used in GARP. Far and by large, <strong>the</strong> primitives in Homer<br />
map one-to-one to <strong>the</strong> primitives in GARP. Whereas <strong>the</strong> functioning <strong>of</strong> most <strong>of</strong> <strong>the</strong> primitives have<br />
been discussed in chapter 4, <strong>the</strong>ir representational equivalent in Homer is summarised briefly:<br />
• <strong>Model</strong> fragments fall into three categories, static, process and agent fragments.<br />
• The condition part <strong>of</strong> model fragments in homer is depicted in red colour in homer, and <strong>the</strong><br />
consequence part is shown in blue colour.<br />
• Supertype model fragments are depicted in green colour.<br />
• Structural relations are defined by using configurations between entities.<br />
• Dependencies between quantities are <strong>the</strong> same as in GARP, and <strong>the</strong>ir direction is visualised by<br />
means <strong>of</strong> arrows. The dependencies map one to one with <strong>the</strong>ir definition in GARP as discussed in<br />
chapter 4.<br />
• Calculations are defined as relations between quantities, and dependencies can be related to <strong>the</strong><br />
sign <strong>of</strong> <strong>the</strong> calculus.<br />
• In Homer, identity relations can be defined in order to specify that two entities refer to <strong>the</strong> same<br />
instance.<br />
Figure 6.3A quantity space in Homer.<br />
Each quantity in Homer has a specified and fixed quantity space, which means that Homer doesn’t<br />
allow <strong>the</strong> use <strong>of</strong> free variables. <strong>An</strong> example <strong>of</strong> a quantity space as defined in Homer is shown in figure<br />
6.3. It shows <strong>the</strong> quantity space <strong>of</strong> substance temperature, and <strong>the</strong> possible values <strong>of</strong> <strong>the</strong> derivative (δ),<br />
zero-plus-max.<br />
6.3.2 <strong>Model</strong> 1<br />
The central processes in <strong>the</strong> water cycle are evaporation condensation, cooling, heating, movement<br />
(flows) and precipitation. In order to classify <strong>the</strong> entities that take pat in <strong>the</strong>se processes, an isahierarchy<br />
is constructed. Figure 6.4 depicts <strong>the</strong> isa-hierarchy for <strong>the</strong> <strong>Model</strong> 1 <strong>of</strong> <strong>the</strong> water cycle.<br />
34
Figure 6.4. Entity hierarchy simple water cycle.<br />
It is assumed here that all <strong>the</strong> objects <strong>of</strong> interest in <strong>the</strong> model are heat exchangeable entities. The<br />
hierarchy contains a substance ‘cloud’, which in principal is <strong>the</strong> same as water and usually not<br />
considered a substance 1 . <strong>An</strong>o<strong>the</strong>r comment is that ‘Air’ is in fact also a kind <strong>of</strong> ‘water storage’, since<br />
‘Air’ may contain water 2 in this model. There are two controller entities in <strong>the</strong> hierarchy, which refer<br />
to types <strong>of</strong> modelling assumptions. When <strong>the</strong> main processes evaporation, condensation,<br />
transportation and precipitation are considered independent from <strong>the</strong>ir context, some <strong>of</strong> <strong>the</strong>se models<br />
already exist. For example, ideas on how condensators or evaporators are modelled qualitatively are<br />
ongoing <strong>the</strong>mes in <strong>the</strong> literature, (Nayak, 1992; Schut & Bredeweg,1993).<br />
A simple model <strong>of</strong> <strong>the</strong> water cycle should minimise <strong>the</strong> use <strong>of</strong> global constraints and focus on <strong>the</strong><br />
individual functioning <strong>of</strong> <strong>the</strong> processes. The model <strong>the</strong>refore leaves implicit <strong>the</strong> causal<br />
interdependencies that <strong>the</strong> processes have. This would be modelled by stating <strong>the</strong> model fragment <strong>of</strong> a<br />
process explicitly in <strong>the</strong> condition <strong>of</strong> a model fragment <strong>of</strong> ano<strong>the</strong>r process.<br />
A model <strong>of</strong> a groundwater flow, or run<strong>of</strong>f, when not including vertical differences and <strong>the</strong> size <strong>of</strong><br />
containers, is that <strong>of</strong> two water storages communicating through a fluid path, see figure 6.4. All<br />
quantity spaces are defined as zero-plus-max, except for <strong>the</strong> flow rate which is zero or plus. For each<br />
water storage, <strong>the</strong> amount <strong>of</strong> water corresponds to <strong>the</strong> level and changes in amount are proportional to<br />
changes in level. The same dependencies hold for level and water pressure. By comparing <strong>the</strong> pressure<br />
on <strong>the</strong> bottom <strong>of</strong> each container, <strong>the</strong> flow rate between <strong>the</strong> containers is calculated.<br />
The model fragment depicted in figure 6.5 is a subtype <strong>of</strong> <strong>the</strong> general fragment liquid flow. In this case<br />
<strong>the</strong> water pressure on <strong>the</strong> left-hand side is greater than <strong>the</strong> water pressure on <strong>the</strong> right hand side. The<br />
amount on <strong>the</strong> left in figure will be influenced negatively as a consequence <strong>of</strong> <strong>the</strong> flow rate becoming<br />
plus. This effect finally propagates back to pressure difference. The positive proportionality between<br />
1 The reason for adding cloud as a substance lies in <strong>the</strong> fact that in this implementation, GARP fails to<br />
instantiate ‘water’ as a new entity when it is already added in <strong>the</strong> system model description as being<br />
contained by a water storage.<br />
35
water pressure and pressure difference means that less water pressure causes <strong>the</strong> pressure difference to<br />
decrease. Note that <strong>the</strong> water pressure on <strong>the</strong> right hand side has a negative proportionality with<br />
pressure difference. In this case increasing water pressure on <strong>the</strong> right side in <strong>the</strong> picture means a<br />
decrease in <strong>the</strong> pressure difference.<br />
Figure 6.5. Groundwater flow in <strong>the</strong> simple water cycle.<br />
The representation <strong>of</strong> <strong>the</strong> process <strong>of</strong> evaporation decreases <strong>the</strong> amount <strong>of</strong> water in liquid phase in<br />
favour <strong>of</strong> <strong>the</strong> amount <strong>of</strong> <strong>the</strong> water in gas phase when <strong>the</strong> temperature <strong>of</strong> water reaches <strong>the</strong> threshold<br />
max, see figure 6.6. The phase <strong>of</strong> <strong>the</strong> substance is usually determined by constraining <strong>the</strong> value <strong>of</strong> for<br />
example temperature or pressure. Such constrain would limit <strong>the</strong> existence <strong>of</strong> vapour to for example<br />
<strong>the</strong> quantity value max for temperature. This constraint is applied here by saying that whenever <strong>the</strong><br />
temperature <strong>of</strong> water reaches max, vapour will develop and its temperature corresponds to that <strong>of</strong><br />
water, which is max.<br />
2 Garp does not support multiple inheritance.<br />
36
Figure 6.6 evaporation in <strong>the</strong> simple water cycle model.<br />
Note that <strong>the</strong> model fragment doesn’t mention <strong>the</strong> occurrence <strong>of</strong> a process <strong>of</strong> heat exchange in <strong>the</strong><br />
condition. The purpose <strong>of</strong> <strong>the</strong> model discussed here is to model <strong>the</strong> processes in such a way that <strong>the</strong><br />
effects <strong>of</strong> <strong>the</strong> context are minimised.<br />
By adding <strong>the</strong> entity ‘Air’, <strong>the</strong> vapour can be contained by air. There is a directed correspondence<br />
between <strong>the</strong> temperature <strong>of</strong> water and vapour. The underlying assumption is that <strong>the</strong> incoming vapour<br />
determines <strong>the</strong> pressure <strong>of</strong> <strong>the</strong> surrounding air, see figure 6.7.<br />
We will assume that evaporation, which results in an increase in amount <strong>of</strong> vapour, has a positive<br />
proportionality with vapour pressure. <strong>An</strong>d vapour pressure has a proportionality with vapour<br />
temperature and air pressure. In addition <strong>the</strong> values <strong>of</strong> vapour pressure and air pressure correspond.<br />
Here <strong>the</strong> influences <strong>of</strong> heat exchanges <strong>of</strong> both vapour and air with o<strong>the</strong>r substances are left out, seeing<br />
<strong>the</strong>m as submissive to <strong>the</strong> main effect <strong>of</strong> evaporation itself.<br />
37
Figure 6.7. Vapour that is contained in <strong>the</strong> air.<br />
The next phase in a water cycle can be that movements <strong>of</strong> <strong>the</strong> air can cause <strong>the</strong> vapour to condense<br />
again. For that to occur, <strong>the</strong>re must exist concurrency <strong>of</strong> cold air (low pressure) and warm air (high<br />
pressure). Two entities <strong>of</strong> air can exchange vapour on <strong>the</strong> basis <strong>of</strong> a pressure difference, which<br />
accounts here for a model <strong>of</strong> transportation <strong>of</strong> air containing vapour, see Figure 6.8. Typical for <strong>the</strong><br />
model here is that <strong>the</strong> temperature <strong>of</strong> <strong>the</strong> vapour corresponds to <strong>the</strong> vapour pressure, in a sense<br />
adjusting to <strong>the</strong> circumstances in <strong>the</strong> air and thus abstracting away <strong>the</strong> actual energy exchange. The<br />
effect is that <strong>the</strong> temperature <strong>of</strong> <strong>the</strong> vapour adjusts to <strong>the</strong> vapour pressure, which in its turn<br />
corresponds to <strong>the</strong> amount <strong>of</strong> vapour. The model thus says for example that vapour coming into cold<br />
air will lose energy to this air by decreasing its temperature. A model <strong>of</strong> a more complex energy<br />
exchange between substances would provide more detail about how <strong>the</strong> vapour and its context <strong>of</strong> air<br />
interact. In this model, <strong>the</strong> transportation model functions as an energy-exchange but leaves <strong>the</strong><br />
process implicit.<br />
38
Figure 6.8. A model for transporting air containing vapour.<br />
The reverse <strong>of</strong> <strong>the</strong> evaporation model is a phase change from gas to liquid when <strong>the</strong> temperature <strong>of</strong> <strong>the</strong><br />
vapour drops below a <strong>the</strong> threshold max. The transportation process causes exactly this effect, and <strong>the</strong><br />
result is that <strong>the</strong> vapour will condense, see Figure 6.9. In order for condensation to occur must be that<br />
<strong>the</strong> amount <strong>of</strong> vapour is greater than zero. The water that is contained by <strong>the</strong> air has no interactions<br />
with o<strong>the</strong>r substances held in <strong>the</strong> air. It cannot exchange energy with o<strong>the</strong>r substances, causing it to for<br />
example evaporate while it is in <strong>the</strong> air.<br />
Figure 6.9. A model <strong>of</strong> condensation.<br />
39
Much in <strong>the</strong> same way, precipitation (Figure 6.10) occurs when <strong>the</strong> amount <strong>of</strong> clouds become greater<br />
than zero. This is <strong>of</strong> course an overly simplified model <strong>of</strong> precipitation. However, <strong>the</strong> current model<br />
focuses on quantities such as temperature and pressure, which do not exclusively influence <strong>the</strong><br />
occurrence <strong>of</strong> precipitation. The precipitation process completes <strong>the</strong> water cycle by influencing <strong>the</strong><br />
amount <strong>of</strong> water in a surface water storage.<br />
Figure 6.10. Precipitation.<br />
The simple model <strong>of</strong> <strong>the</strong> water cycle leaves out quite some detail. Next to leaving out many <strong>of</strong> <strong>the</strong><br />
energy exchanges that occur between various substances, some o<strong>the</strong>r features have not been included.<br />
For example, rivers usually flow to <strong>the</strong> sea because <strong>the</strong>re is a certain vertical difference between <strong>the</strong><br />
sea and river, which triggers a downward stream. Also <strong>the</strong> vertical differences in temperature in <strong>the</strong><br />
atmosphere when modelled, add more detail to <strong>the</strong> model. Vertical differences are partly responsible<br />
for energy exchanges between substances and <strong>the</strong>ir context, for example <strong>the</strong> atmosphere. In general,<br />
<strong>the</strong> behaviour <strong>of</strong> air layers and <strong>the</strong> various substances contained by it can be modelled in more detail<br />
as well. More detail can also be added by making explicit <strong>the</strong> interdependencies that hold between<br />
various processes. In <strong>Model</strong> 1, <strong>the</strong> conditions <strong>of</strong> processes do not assume <strong>the</strong> existence and interaction<br />
with o<strong>the</strong>r parts <strong>of</strong> <strong>the</strong> system. The next model puts more focus on <strong>the</strong> integral functioning <strong>of</strong> <strong>the</strong><br />
system<br />
6.3.3 <strong>Model</strong> 2<br />
<strong>Model</strong> 2 introduces more detail in describing <strong>the</strong> water cycle. It focuses on making explicit <strong>the</strong><br />
functioning <strong>of</strong> processes in interaction with o<strong>the</strong>r processes. It also defines new types <strong>of</strong> processes that<br />
are left implicit or left out in model 1.<br />
40
In <strong>the</strong> isa hierarchy <strong>of</strong> <strong>the</strong> model in figure 6.11, an initial division is made between physical object,<br />
heat-exchangeable object, status and assumptions. Status represents a mode <strong>of</strong> existence, like a phase<br />
<strong>of</strong> a substance. Subtypes <strong>of</strong> physical object are objects such as containers, substances and paths, are<br />
heat exchangeable entities. Attributes like stiffness, size, openness and phase are subtypes <strong>of</strong> status;<br />
stiffness is a measure to distinguish between pressure areas and surface storages (elastic or rigid),<br />
phase refers to <strong>the</strong> physical phase (liquid, gas or solid)<br />
All substances have a common name and an attribute that determines <strong>the</strong> phase. Creating a piece <strong>of</strong><br />
water is done by instantiating water from substance and liquid from phase. For instance we give <strong>the</strong><br />
entity water <strong>the</strong> variable name seawater with an attribute has_phase “liquid”. Using this approach it is<br />
<strong>the</strong> same as saying that for example <strong>the</strong> air we brea<strong>the</strong> is O2 in gas phase.<br />
Figure 6.11. Entity hierarchy.<br />
Several kinds <strong>of</strong> physical entities are distinguished. In <strong>the</strong> following, definitions on individuals such as<br />
substances, containers, contained substances, paths and portals will be discussed. When<br />
individualising objects or concepts, one <strong>of</strong> <strong>the</strong> criteria must be <strong>the</strong> spatio-temporal continuity <strong>of</strong> <strong>the</strong><br />
objects (Hayes, 1978). For substances such as liquids <strong>the</strong> individualisation can be done in an assembly<br />
with containers that contain it, similar to <strong>the</strong> contained stuff ontology.<br />
Substances and containers exist in separate views as well as in assemblies <strong>of</strong> both, <strong>the</strong> contained<br />
substances. This is in fact slightly different than <strong>the</strong> contained stuff ontology, where no liquids can be<br />
considered that are not contained in a container. The model in itself has no knowledge to reason about<br />
substances, it can only reason about <strong>the</strong> behaviour <strong>of</strong> contained substances. Containers are considered<br />
with respect to <strong>the</strong>ir volume, heat, pressure, stiffness and openness. In addition, containers may be<br />
equipped with a portal.<br />
There is a substantial difference between containers that are modelled open or closed, this has to do<br />
with exposure to <strong>the</strong> atmosphere, see also (Collins & Forbus 1989, p16.). In case <strong>of</strong> a closed container,<br />
<strong>the</strong> exposure to <strong>the</strong> atmosphere doesn't influence <strong>the</strong> quantities inside <strong>the</strong> container, such as air<br />
pressure. In <strong>the</strong> current scope, we assume <strong>the</strong> existence <strong>of</strong> closed containers with elastic features,<br />
<strong>the</strong>se mapping to for example a pressure area in <strong>the</strong> atmosphere. In principle this means that elastic<br />
41
containers are subject to exposure in <strong>the</strong> atmosphere and as a consequence, <strong>the</strong> pressure inside <strong>the</strong><br />
container may be influenced as well (reaction force). In <strong>the</strong> contained stuff ontology, various<br />
properties <strong>of</strong> individuals, such as pressure or temperature exist when for example a container, a<br />
substance and its phase are assembled into a contained stuff. A contained substance is regarded as an<br />
individual when <strong>the</strong> following function applies:<br />
contained stuffs –→ substance x phase x container.<br />
Substances always have <strong>the</strong> quantity heat and amount. By integrating substance and phase in one<br />
view, we create <strong>the</strong> specific states <strong>of</strong> matter such as gas, liquid or solid, accompanied with necessary<br />
quantity conditions using <strong>the</strong> quantity space for substance-temperature: above vapourpoint-between<br />
vapour and freezepoint-freezepoint-below freezepoint-absolute nil. Table 6.1 gives an overview <strong>of</strong> <strong>the</strong><br />
quantity conditions for different states <strong>of</strong> matter.<br />
phase(substance): quantity condition:<br />
solid temperature(substance) ≤ freezing-point(temperature)<br />
liquid temperature(substance) ≤ vapour-point(temperature),<br />
temperature(substance) ≥ freezepoint(temperature)<br />
gas temperature(substance) ≥ vapour-point(temperature)<br />
Table 6.1. Quantity conditions for substances.<br />
The energy exchanges in <strong>the</strong> system are modelled between substances. For instance in <strong>the</strong> model, <strong>the</strong><br />
sea (open contained liquid) and a pressure area (closed elastic contained gas) are able to exchange<br />
heat. In order to create a flow rate, a path is introduced between heat connected substances. The<br />
quantity space for heat is limited to <strong>the</strong> interval plus for all entities.<br />
<strong>An</strong> ‘contained substance’ or ‘open contained substance’ does not introduce <strong>the</strong> quantity volume. The<br />
rationale is that in <strong>the</strong> current perspective, <strong>the</strong> added value <strong>of</strong> introducing this quantity is neglectable<br />
in comparison to <strong>the</strong> impact that a ‘closed contained gas’ has on <strong>the</strong> quantity volume <strong>of</strong> a container.<br />
The model fragment for ‘closed contained gas’, depicted in figure 6.12. It is a subtype <strong>of</strong> <strong>the</strong> more<br />
general ‘contained substance, and has as condition <strong>the</strong> existence <strong>of</strong> a closed container and a gas. A<br />
‘closed contained gas’ introduces <strong>the</strong> quantities volume and pressure for <strong>the</strong> container, and <strong>the</strong><br />
following dependencies:<br />
Amount<br />
Heat<br />
Pressure<br />
Volume<br />
α Q+<br />
α Q+<br />
α Q+<br />
α Q-<br />
42<br />
Pressure<br />
Pressure<br />
Temperature<br />
Pressure
Figure 6.12. <strong>Model</strong> fragment <strong>of</strong> “closed contained gas”.<br />
The dependencies for closed contained gasses follow <strong>the</strong> usual statements in physics textbooks, see<br />
also (Forbus, 1984). The quantity spaces for <strong>the</strong>se parameters are all <strong>of</strong> <strong>the</strong> type zero-plus-max.<br />
A subtype model fragment <strong>of</strong> <strong>the</strong> ‘closed contained gas’ fragment controls <strong>the</strong> relations between <strong>the</strong><br />
amount <strong>of</strong> substance inside <strong>the</strong> container and <strong>the</strong> volume and pressure <strong>of</strong> <strong>the</strong> container itself. The<br />
fragment ‘contained gas volume pressure details’ says that when <strong>the</strong> amount <strong>of</strong> substance is greater<br />
than zero, <strong>the</strong> following undirected qualitative correspondences exist:<br />
Amount Qcorr Volume<br />
Pressure Qcorr Volume<br />
6.3.3.1 <strong>Multiple</strong> gasses in a container<br />
Following standard definitions <strong>of</strong> containers and contained spaces, <strong>the</strong> former is normally a rigid<br />
object and <strong>the</strong> latter is in a relation to <strong>the</strong> container and should include a 3D-space with contiguous<br />
rigid boundaries surrounding it (Hayes, 1978). In our model decisions were made that are somewhat<br />
contrary to <strong>the</strong>se definitions and <strong>the</strong>refore must be discussed in more detail. The core <strong>of</strong> <strong>the</strong> discussion<br />
lies in <strong>the</strong> definition <strong>of</strong> a pressure area. A pressure area should form a contained space, distinguishing<br />
it from o<strong>the</strong>r contained spaces such as o<strong>the</strong>r pressure areas in a bigger contained space, <strong>the</strong><br />
atmosphere. But <strong>the</strong>re are <strong>of</strong> course no visible boundaries between pressure area in <strong>the</strong> sky, let alone<br />
<strong>the</strong>y are rigid and physical. The reason why <strong>the</strong>re does exist some correspondence between a pressure<br />
area and a container is that <strong>the</strong> behaviour manifested by certain volumes <strong>of</strong> stuffs in <strong>the</strong> atmosphere is<br />
43
esembled well by a balloon filled with <strong>the</strong> same stuff in <strong>the</strong> atmosphere. As a formal equivalent <strong>the</strong><br />
choice is made to distinguish between elastic containers and rigid containers for stuffs contained in <strong>the</strong><br />
atmosphere and surface respectively. A pressure area is a closed elastic container containing multiple<br />
substances, and is itself contained in <strong>the</strong> atmosphere.<br />
Until this point we have considered only separately contained substances. When considering what is<br />
thus called a pressure area we must describe how different substances can be jointly contained in <strong>the</strong><br />
same container, thus describing <strong>the</strong> causal dependencies between shared parameters <strong>of</strong> <strong>the</strong> substances<br />
in <strong>the</strong> container such as pressure, volume and heat. The first problem is how to derive <strong>the</strong> value for<br />
pressure. Let’s take <strong>the</strong> analogy with a balloon again. There are two decisive distinctions between a<br />
pressure area and a balloon. First <strong>of</strong> all, a balloon has actual physical boundaries where a pressure area<br />
has none. Secondly, although it’s boundaries are elastic <strong>the</strong>re exists a limit to <strong>the</strong> amount <strong>of</strong> 3D-space<br />
a balloon can fill and a limit to <strong>the</strong> pressure when <strong>the</strong> balloon has expanded to its maximum. A<br />
pressure area has no physical boundaries and is <strong>the</strong>refore virtually unconstrained in space. In <strong>the</strong><br />
model <strong>the</strong> behaviour <strong>of</strong> only one or two pressure areas is considered at a time. This means that o<strong>the</strong>r<br />
pressure area’s in <strong>the</strong> atmosphere will not be able to sufficiently constrain <strong>the</strong> volume <strong>of</strong> <strong>the</strong> area.<br />
Consequently, <strong>the</strong> pressure will not be able to increase significantly within <strong>the</strong> area. The only way to<br />
find a constraint mechanism for pressure is to assign a fixed value to <strong>the</strong> volume for <strong>the</strong> pressure area.<br />
Therefore an agent model is introduced, which constrains <strong>the</strong> volume <strong>of</strong> a pressure area to plus and <strong>the</strong><br />
derivative to zero.<br />
The fragment ‘assembly pressure area’ defines that a pressure area contains two types <strong>of</strong> closed<br />
contained gas, and one closed contained liquid: air pressure area, vapour and water pressure area. The<br />
overall pressure <strong>of</strong> <strong>the</strong> assembly is influenced by changes in amount <strong>of</strong> substances. The quantity<br />
pressure in its turn has a positive qualitative proportionality with <strong>the</strong> temperature <strong>of</strong> <strong>the</strong> substances.<br />
This makes <strong>the</strong> circle complete: any increase <strong>of</strong> amount <strong>of</strong> a substance leads to increasing pressure in<br />
<strong>the</strong> assembly and this leads to an increase in temperature <strong>of</strong> <strong>the</strong> remaining substances in <strong>the</strong> pressure<br />
area. In <strong>the</strong> meanwhile, <strong>the</strong> substances that are contained in <strong>the</strong> pressure area will exchange energy<br />
through pre-defined heat paths.<br />
The fragment ‘assembly atmosphere’ is a closed contained gas that contains ano<strong>the</strong>r assembly, <strong>the</strong><br />
pressure area. Because <strong>of</strong> its containment in <strong>the</strong> atmosphere, <strong>the</strong> vertical position <strong>of</strong> <strong>the</strong> pressure area<br />
is added by introducing <strong>the</strong> quantity altitude. It uses <strong>the</strong> quantity space atmosphere levels with values<br />
high-middle-low. In figure 6.13, <strong>the</strong> structure (entities and quantities) <strong>of</strong> <strong>the</strong> model fragment<br />
“assembly atmosphere” is depicted as it is instantiated in VisiGarp. All relations and dependencies are<br />
left out <strong>of</strong> <strong>the</strong> picture.<br />
44
Figure 6.13. Entities and quantities in “assembly atmosphere”.<br />
By specifying identity relations in Homer, <strong>the</strong> system find that <strong>the</strong> containers in which <strong>the</strong> substances<br />
<strong>of</strong> ‘assembly pressure area’ are contained are all referring to <strong>the</strong> same instance <strong>of</strong> one container. In <strong>the</strong><br />
figure this can be seen by noticing that <strong>the</strong>re exist only two quantities volume (1 and 2) and pressure<br />
(1 and 2). The ‘assembly atmosphere’ thus consists <strong>of</strong> two containers and four substances.<br />
6.3.3.2 vertical differences on <strong>the</strong> surface<br />
In <strong>the</strong> implementation, surface layers are represented as open contained stuffs. These can be rivers,<br />
lakes or oceans. Although <strong>the</strong>y are in constant flux, we will thus consider <strong>the</strong>m as individuals. The<br />
container refers to <strong>the</strong> 3-dimensional space, which can have a portal at a certain position. The purpose<br />
<strong>of</strong> this representation is to reason about vertical differences <strong>of</strong> liquids that propagate to pressure<br />
differences on each side <strong>of</strong> a path that is connected to portal <strong>of</strong> containers. The different portal<br />
positions require different quantity spaces to be assigned to <strong>the</strong> level and amount <strong>of</strong> substance in a<br />
container, see figure 6.12. Assembling different types <strong>of</strong> contained stuffs makes it possible to simulate<br />
water features such as a river (which typically runs above sea level) emptying in <strong>the</strong> ocean.<br />
In <strong>the</strong> approach taken here, <strong>the</strong> pressure values are assigned for every possible configuration <strong>of</strong> portal<br />
and level. With <strong>the</strong> idea outlined above we can assume a container geometry resembling <strong>the</strong> one in<br />
figure 6.14.<br />
45
Figure 6.14. Container geometry for surfaces.<br />
Before processes involving multiple objects can be described, <strong>the</strong> connectivity <strong>of</strong> <strong>the</strong> objects involved<br />
must be discussed, yielding a structural description <strong>of</strong> <strong>the</strong> situation at hand. The structural description<br />
should provide a set <strong>of</strong> constraints that allow <strong>the</strong> simulator to make certain inference about <strong>the</strong><br />
difference in pressure on both portals.<br />
The critical factor in modelling and simulating <strong>the</strong> flows that occur between different surfaces on land<br />
will be <strong>the</strong> level <strong>of</strong> <strong>the</strong> column <strong>of</strong> water above a path that connects one container to ano<strong>the</strong>r. Paths are<br />
connected to containers through portals and portals can be placed ei<strong>the</strong>r on <strong>the</strong> bottom, <strong>the</strong> middle or<br />
<strong>the</strong> top <strong>of</strong> a container. The level <strong>of</strong> <strong>the</strong> liquid relative to <strong>the</strong> position <strong>of</strong> <strong>the</strong> portal propagates to<br />
pressure on <strong>the</strong> portal. For example in a container with a portal placed in <strong>the</strong> middle and a level equal<br />
to max, <strong>the</strong> pressure on <strong>the</strong> portal must be plus. In <strong>the</strong> approach taken here, <strong>the</strong> pressure values are<br />
assigned for every possible configuration <strong>of</strong> portal and level, see also figure 6.14, which depicts one <strong>of</strong><br />
<strong>the</strong> nine possible configurations. The two contained liquids in <strong>the</strong> picture have <strong>the</strong> same level, or more<br />
precise, <strong>the</strong> same amount <strong>of</strong> liquid when <strong>the</strong> two containers are assumed to have equal sizes. The<br />
quantity spaces level liquid and amount describe <strong>the</strong> various levels and amounts <strong>of</strong> liquid relative to<br />
<strong>the</strong> portal, which itself can thus have variable height. Open contained liquids with a portal at <strong>the</strong> top or<br />
bottom <strong>of</strong> <strong>the</strong> container use different quantity spaces for level and amount than open contained liquids<br />
with a portal at <strong>the</strong> middle (table 6.2).<br />
46
Structure Substance level quantity space Substance amount quantity space<br />
Open contained liquid<br />
With top portal<br />
Open contained liquid<br />
with bottom portal<br />
Open contained liquid<br />
with middle portal<br />
bottom container – plus – top container zero – plus – max<br />
bottom container – plus – top container zero – plus – max<br />
bottom container – below portal – portal<br />
place – above portal – top container<br />
Table 6.2. Possible quantity spaces for Open contained liquids.<br />
47<br />
zero – below portal – portal place –<br />
above portal – max<br />
The portal pressure in a open contained liquid with a top portal can never be positive, since <strong>the</strong><br />
substance cannot form a column above <strong>the</strong> top <strong>of</strong> <strong>the</strong> container. The quantities amount and level <strong>of</strong><br />
substance are qualitatively proportional and correspond. For open contained liquids with a bottom<br />
portal all three quantities, substance level, substance amount and portal pressure are qualitatively<br />
proportional. Figure 6.15 shows a container with a portal in <strong>the</strong> middle.<br />
Figure 6.15. <strong>An</strong> open contained liquid with portal in <strong>the</strong> middle.
In <strong>the</strong> figure 6.15, all possible values for level have a corresponding value with portal pressure. The<br />
dependencies for <strong>the</strong> quantity portal pressure can be made more general when more information is<br />
available. For example when we know that <strong>the</strong> level is above <strong>the</strong> position <strong>of</strong> <strong>the</strong> portal, we can add<br />
information by saying that <strong>the</strong> pressure on <strong>the</strong> portal increases as <strong>the</strong> level rises. Conversely, when <strong>the</strong><br />
level drops below <strong>the</strong> position <strong>of</strong> <strong>the</strong> portal, <strong>the</strong> pressure on <strong>the</strong> portal will always be zero. These<br />
issues are defined in two subtype model fragments <strong>of</strong> ‘open contained liquid middle portal’.<br />
The idea is that when two containers are connected by a path, and <strong>the</strong> level in one <strong>of</strong> <strong>the</strong> containers<br />
increases, <strong>the</strong>n so will <strong>the</strong> portal pressure <strong>of</strong> <strong>the</strong> container and flow can occur. This flow has to restore<br />
<strong>the</strong> equilibrium in pressure on both portals. The use <strong>of</strong> portals with contained liquids and o<strong>the</strong>r issues<br />
concerning possible configurations <strong>of</strong> fluid flows are reflected upon in (Collins & Forbus, 1987).<br />
Somewhat different from <strong>the</strong> purpose <strong>of</strong> using portals by Collins and Forbus, who use portals for<br />
resolving ambiguity with causality within fluid paths, here portals are used mainly to produce<br />
constraints that account for vertical distinctions between surfaces. Several types <strong>of</strong> paths can exist in<br />
connections. We distinguish path, fluid path, gas path and heat path, where <strong>the</strong> first is an unspecified<br />
super type <strong>of</strong> <strong>the</strong> o<strong>the</strong>rs. Paths may connect only two distinct containers. In specific cases, like <strong>the</strong><br />
surface flow, where specific position <strong>of</strong> a container is relevant and thus <strong>the</strong> direction <strong>of</strong> <strong>the</strong> flow,<br />
specific attributes such as left hand side or right hand side are added to <strong>the</strong> paths. Figure 6.16 shows a<br />
qualitative structural description <strong>of</strong> <strong>the</strong> connection <strong>of</strong> surface containers.<br />
Figure 6.16. Generic definition <strong>of</strong> a portal connection.<br />
The figure shows container A with portal A and container B with portal B being connected with a fluid<br />
path which has portal A at its left side and portal B at its right side. The description is generic in <strong>the</strong><br />
sense that no information is needed concerning <strong>the</strong> positions <strong>of</strong> <strong>the</strong> portals. This information is needed<br />
when finally in figure 6.16, a specific portal connection is made which requires two distinctly named<br />
contained stuffs to participate.<br />
48
Figure 6.17. Portal connection: bottom to middle.<br />
The quantities level <strong>of</strong> two substances always have a relation describing how <strong>the</strong> level <strong>of</strong> one<br />
substance is relative to <strong>the</strong> level <strong>of</strong> <strong>the</strong> second substance. In figure 6.17, <strong>the</strong> consequence <strong>of</strong> a open<br />
container with bottom portal being connected to and open container with middle portal is that <strong>the</strong><br />
values bottom container (levelA) and portal place (levelB) are equal. The result <strong>of</strong> this method is that<br />
<strong>the</strong> simulator gets nearly all <strong>the</strong> information it needs from <strong>the</strong> structural description and does not have<br />
to calculate for example <strong>the</strong> pressure by inferring <strong>the</strong> difference between <strong>the</strong> level and <strong>the</strong> portal<br />
position. The current approach results in a structural description <strong>of</strong> every possible configuration <strong>of</strong><br />
containers and portal positions. All nine possible configurations and <strong>the</strong>ir specific equality relations<br />
for level are given in table 6.3.<br />
49
Portal connection Relation<br />
1. top to top equal( top_container( LevelA ), top_container( LevelB ) )<br />
2. top to middle equal( top_container( LevelA ), portal_place( LevelB ) )<br />
3. top to bottom equal( top_container( LevelA ), bottom_container( LevelB ) )<br />
4. middle to top equal( portal_place( LevelA ), top_container( LevelB ) )<br />
5. middle to middle equal( portal_place( LevelA ), portal_place( LevelB ) )<br />
6. middle to bottom equal( portal_place( LevelA ), bottom_container( LevelB ) )<br />
7. bottom to top equal( bottom_container( LevelA ), top_container( LevelB ) )<br />
8. bottom to middle equal( bottom_container( LevelA ), portal_place( LevelB ) )<br />
9. bottom to bottom equal( bottom_container( LevelA ), bottom_container( LevelB )<br />
Table 6.3. Relative position <strong>of</strong> level for two contained substances.<br />
In order to create a less abstract representation <strong>of</strong> surface water flows, <strong>the</strong> model can be elaborated<br />
with for example <strong>the</strong> notion <strong>of</strong> permeability <strong>of</strong> surface containers. With this idea, <strong>the</strong> boundaries <strong>of</strong> <strong>the</strong><br />
containers must be assumed to be impermeable, o<strong>the</strong>rwise portals should be assigned. This serves as a<br />
distinction between permeable and impermeable ground layers. Since soil itself may hold liquid<br />
between its pieces <strong>of</strong> stuff, a container with no portals suits <strong>the</strong> idea, hence it allows no water to flow<br />
in or out. The level <strong>of</strong> <strong>the</strong> liquid in <strong>the</strong> container resembles <strong>the</strong> groundwater table. Similarly <strong>the</strong> space<br />
above <strong>the</strong> water table compares to <strong>the</strong> zone <strong>of</strong> percolation, which is <strong>the</strong> (non-) permeable layer.<br />
Consequently, for <strong>the</strong> water that reaches soil surfaces as a result <strong>of</strong> precipitation, three alternative<br />
kinds <strong>of</strong> behaviour are possible. In <strong>the</strong> first case it can infiltrate <strong>the</strong> soil given that <strong>the</strong> soil layer is<br />
permeable, e.g. a portal and a path exist. In <strong>the</strong> second case, it is possible that <strong>the</strong> soil is permeable but<br />
<strong>the</strong> level <strong>of</strong> <strong>the</strong> water table is maximal, which means that any fur<strong>the</strong>r infiltration will result in flooding,<br />
run<strong>of</strong>f overland to a river. Third, it is possible that <strong>the</strong> soil surface is impermeable, which means that it<br />
runs over <strong>the</strong> surface as well. A quite nice representation to this in terms <strong>of</strong> QR was thought <strong>of</strong> during<br />
a course in model based reasoning about this particular domain. It is possible to model some kind <strong>of</strong><br />
intermediate storage containers, resembling puddles. Puddles can be modelled relatively<br />
straightforward as open contained stuffs. This elaboration would allow a specification in <strong>the</strong> model<br />
between land and rivers or seas.<br />
6.3.3.3 Processes<br />
As mentioned before, energy exchanges are modelled between substances in <strong>the</strong> model. With a<br />
difference in temperature as condition, heat is exchanged with a rate that is equal to <strong>the</strong> difference in<br />
pressure. The “heat flow” fragment has a general top level and three specialisations, which model <strong>the</strong><br />
direction <strong>of</strong> <strong>the</strong> flow. Because <strong>the</strong> flow rate has a quantity space containing min-zero-plus, <strong>the</strong><br />
influences specified in <strong>the</strong> general fragment apply to each direction <strong>of</strong> heat flow. In figure 6.18, <strong>the</strong><br />
subtype “heat flow negative” is depicted. Because <strong>the</strong> flow rate is min, <strong>the</strong> amount <strong>of</strong> heat <strong>of</strong> <strong>the</strong><br />
50
source will increase and <strong>the</strong> amount <strong>of</strong> heat in <strong>the</strong> destination will decrease. The sign <strong>of</strong> <strong>the</strong> influence<br />
relation between quantities is thus opposite to its actual influence when <strong>the</strong> value <strong>of</strong> <strong>the</strong> influencing<br />
quantity is below zero.<br />
Figure 6.18. Heat flow negative.<br />
Only substances that are connected to ano<strong>the</strong>r substance via a pre-specified heat path can engage in<br />
energy exchange. The temperature <strong>of</strong> <strong>the</strong> source is positively proportional to <strong>the</strong> flow rate, and <strong>the</strong><br />
temperature <strong>of</strong> <strong>the</strong> destination is negatively proportional to <strong>the</strong> flow rate. Changes in temperature thus<br />
may affect <strong>the</strong> intensity <strong>of</strong> energy exchange.<br />
Conditional on <strong>the</strong> existence <strong>of</strong> a heat flow between <strong>the</strong> sun and <strong>the</strong> seawater, water may start to<br />
evaporate when <strong>the</strong> temperature <strong>of</strong> <strong>the</strong> water is at vapour point and its derivative is greater or equal<br />
than plus. Different than for example <strong>the</strong> model <strong>of</strong> evaporation in <strong>Model</strong> I, <strong>the</strong> process is dependent on<br />
<strong>the</strong> occurrence <strong>of</strong> <strong>the</strong> process <strong>of</strong> heat flow. The flow rate between <strong>the</strong> heat-exchanging entities also<br />
specifies <strong>the</strong> flow rate <strong>of</strong> evaporation. As you can see in figure 6.19, <strong>the</strong> model fragment “evaporation”<br />
also has consequences for <strong>the</strong> quantity heat <strong>of</strong> both substances. When <strong>the</strong> amount <strong>of</strong> a substance<br />
changes, <strong>the</strong> heat <strong>of</strong> <strong>the</strong> substances changes in <strong>the</strong> same direction. The model is specific for a heat flow<br />
between at least one designated substance, because <strong>the</strong>re must exist a positive heat flow between<br />
seawater and ano<strong>the</strong>r substance.<br />
51
Figure 6.19. Evaporation occurs when <strong>the</strong>re is heat flow.<br />
The atmosphere and a pressure area can exchange heat by a flow rate that triggers on a difference in<br />
temperature between <strong>the</strong> air <strong>of</strong> <strong>the</strong> atmosphere and <strong>the</strong> air <strong>of</strong> <strong>the</strong> pressure area, and <strong>the</strong> existance <strong>of</strong> a<br />
heath path connecting both. A consequence <strong>of</strong> <strong>the</strong> energy exchange is that, given this heat flow and a<br />
pressure difference between <strong>the</strong> two, <strong>the</strong> pressure area may engage a vertical movement. Three types<br />
<strong>of</strong> movement are possible:<br />
1. Pressure-Pressure area > Pressure Atmosphere: upward movement<br />
2. Pressure-Pressure area = Pressure Atmosphere: no movement<br />
3. Pressure-Pressure area < Pressure Atmosphere: downward movement<br />
Because <strong>the</strong> pressure decreases in <strong>the</strong> atmosphere at higher altitudes, <strong>the</strong> differences in pressure<br />
between <strong>the</strong> atmosphere and <strong>the</strong> pressure area will change depending on <strong>the</strong> altitude <strong>of</strong> <strong>the</strong> pressure<br />
area. The pressure <strong>of</strong> <strong>the</strong> atmosphere decreases at higher altitudes. In order to calculate <strong>the</strong> difference<br />
between <strong>the</strong> pressure <strong>of</strong> <strong>the</strong> atmosphere, which should be specific to altitude, and <strong>the</strong> pressure <strong>of</strong> <strong>the</strong><br />
pressure area, it would be necessary to define three different quantities for pressure in <strong>the</strong> atmosphere,<br />
namely one quantity per possible value for altitude. This would result in a qualitative representation <strong>of</strong><br />
an equation <strong>of</strong> <strong>the</strong> type ∆(PressurePA – ∆ PressureATM).<br />
52
Adding this representation implies a thorough specialisation <strong>of</strong> <strong>the</strong> more general “closed contained<br />
gas” and most <strong>of</strong> <strong>the</strong> process fragments in <strong>the</strong> model.<br />
Instead, when any pressure movement occurs, a quantity vertical pressure gradient is introduced.<br />
Figure 6.20 depicts <strong>the</strong> process <strong>of</strong> pressure area movement. The idea is that depending on <strong>the</strong><br />
movement <strong>of</strong> <strong>the</strong> pressure area, <strong>the</strong> atmosphere influences <strong>the</strong> pressure <strong>of</strong> <strong>the</strong> pressure area in opposite<br />
direction. For example upward movement causes a negative influence on <strong>the</strong> pressure <strong>of</strong> <strong>the</strong> pressure<br />
area, and vice versa. Because <strong>the</strong> influence corresponds to <strong>the</strong> movement <strong>of</strong> <strong>the</strong> pressure area, <strong>the</strong><br />
energy exchange and its relation to altitude is left implicit.<br />
Figure 6.20. Vertical movement <strong>of</strong> a pressure area.<br />
The assumption discussed that <strong>the</strong> volume <strong>of</strong> a pressure area always remains unchanged has <strong>the</strong> effect<br />
that <strong>the</strong> pressure area moves upward and continues to exchange energy with <strong>the</strong> atmosphere. In <strong>the</strong><br />
meantime, <strong>the</strong> substances in <strong>the</strong> pressure area exchange heat by <strong>the</strong> same type <strong>of</strong> process. Vapour for<br />
example will start to exchange energy with <strong>the</strong> air <strong>of</strong> <strong>the</strong> pressure area once <strong>the</strong> latter cools down. As a<br />
consequence, <strong>the</strong> temperature <strong>of</strong> vapour will decrease to <strong>the</strong> lowest threshold for a substance to be in<br />
53
gas phase. When <strong>the</strong> temperature <strong>of</strong> vapour decreases to vapour point, and its derivative is smaller<br />
than zero, condensation occurs. Figure 6.21 depicts this process. It has largely <strong>the</strong> same structure as<br />
<strong>the</strong> model fragment “evaporation”. Again, <strong>the</strong> process is explicitly dependent on <strong>the</strong> occurrence <strong>of</strong> a<br />
heat exchange, here between air and water in a pressure area. Also <strong>the</strong> amount <strong>of</strong> energy <strong>of</strong> <strong>the</strong><br />
substances is changed according to <strong>the</strong> changes in amount <strong>of</strong> substances.<br />
Figure 6.21. Condensation.<br />
Only a certain amount <strong>of</strong> water can be kept in a pressure area. When <strong>the</strong> amount <strong>of</strong> water reaches max,<br />
<strong>the</strong> air is saturated with water. The model fragment “precipitation” in figure 6.22, is modelled as<br />
being dependent on <strong>the</strong> occurrence <strong>of</strong> <strong>the</strong> process condensation. When <strong>the</strong> amount <strong>of</strong> water is max, and<br />
<strong>the</strong> amount increases as a consequence <strong>of</strong> condensation, <strong>the</strong> flow rate through <strong>the</strong> liquid path<br />
corresponds to <strong>the</strong> flow rate <strong>of</strong> condensation. Note that in this model, <strong>the</strong> quantity heat <strong>of</strong> <strong>the</strong> seawater<br />
is not influenced by <strong>the</strong> increase <strong>of</strong> <strong>the</strong> amount <strong>the</strong> water.<br />
54
Figure 6.22. The model fragment “precipitation”.<br />
By modelling processes as being explicitly interdependent on <strong>the</strong> occurrence <strong>of</strong> some o<strong>the</strong>r process a<br />
rigid and complex causal model is constructed. The core <strong>of</strong> <strong>the</strong> causal model is <strong>the</strong> energy exchange<br />
between substances. The scenario used for <strong>the</strong> model defines several heat paths, listed in table 6.4:<br />
Source substance<br />
Heat path Destination substance<br />
Hydrogen from Sun to Sea <strong>Water</strong>(sea)<br />
Vapour from Vapour to Air(Pressure area) Air(Pressure area)<br />
Vapour from Vapour to <strong>Water</strong>(Pressure area) <strong>Water</strong>(Pressure area)<br />
<strong>Water</strong>(Pressure area) from <strong>Water</strong>(Pressure area) to Air(Pressure area) Air(Pressure area)<br />
Air(Pressure area) from Air(Pressure area) to Air(Atmosphere) Air(Atmosphere)<br />
Table 6.4. Possible heat paths between substances.<br />
The processes in <strong>Model</strong> 2 all have an (indirect) effect on <strong>the</strong> energy balance between substances. The<br />
consequence <strong>of</strong> which is that <strong>the</strong> heat flow occurs continuously when <strong>the</strong> model is simulated. In <strong>Model</strong><br />
1, <strong>the</strong>re are no explicit interdependencies modelled for processes. The occurrence <strong>of</strong> (conditional)<br />
processes is left implicit in <strong>Model</strong> 1. However when simulated, <strong>the</strong> resulting simulation will follow a<br />
similar causal path <strong>of</strong> behaviour, because <strong>the</strong> conditions <strong>of</strong> <strong>the</strong> processes in model 1 cannot be<br />
satisfied without <strong>the</strong> occurrence <strong>of</strong> o<strong>the</strong>r processes.<br />
55
7. Simulation <strong>of</strong> <strong>the</strong> water cycle<br />
In this chapter, <strong>the</strong> current status <strong>of</strong> <strong>the</strong> qualitative simulation <strong>of</strong> <strong>the</strong> water cycle is described.<br />
Currently <strong>the</strong>re exist three runnable models. <strong>Water</strong> cycle 1 (<strong>Model</strong> 1), <strong>Water</strong> cycle 2 (<strong>Model</strong> 2), and a<br />
running model <strong>of</strong> vertical differences <strong>of</strong> contained substances. As <strong>of</strong> yet <strong>the</strong> different models have not<br />
been combined into an overall simulation model, which could for example support <strong>the</strong> idea <strong>of</strong> model<br />
progression by means <strong>of</strong> different scenario’s that vary in <strong>the</strong>ir focus, perspective and detail. Each<br />
single simulation will be reflected upon next.<br />
7.1 Simulating vertical differences<br />
The simulation will be illustrated by an example simulation <strong>of</strong> two containers, one with a top portal<br />
(left side), and one with a middle portal (right side). The scenario is depicted in figure 7.1.<br />
Figure 7.1. Scenario for top-middle containers.<br />
The amount on <strong>the</strong> left-hand side is zero and <strong>the</strong> portal pressure <strong>of</strong> a container with a top portal is zero<br />
by definition. <strong>An</strong>y value above Portal place for <strong>the</strong> amount in <strong>the</strong> container on <strong>the</strong> right-hand side<br />
would render an inequality between both portal pressures. The simulator finds this in <strong>the</strong> first state.<br />
56
The inequality triggers a liquid flow process by instantiating <strong>the</strong> fragment “flow from right-hand-side<br />
to left-hand-side”. The model fragment flow from right-hand-side to left-hand-side is depicted below<br />
in figure 7.2, generated using VISIGARP. Because <strong>the</strong> model fragment is a subtype <strong>of</strong> <strong>the</strong> more<br />
general ‘liquid flow’, only <strong>the</strong> inequality is highlighted. The dependencies between level and pressure<br />
are instantiated by <strong>the</strong> model fragments for containers with a specific portal, described in paragraph<br />
6.4.3.2.<br />
Figure 7.2. Flow from right-hand-side to left-hand-side.<br />
Moving on to Figure 7.3, <strong>the</strong> green circles are <strong>the</strong> possible states resulting from <strong>the</strong> predecessor(s).<br />
Black circles are terminated states with no successors. In state 1, an inequality is found and <strong>the</strong> liquid<br />
flow fragment is triggered. The simulator finds <strong>the</strong> same inequality discussed above in state two, and<br />
after that only in state 5 and not in states 3 and 4(figure 6.24a). In figure 6.24b, we see that in fact state<br />
five leads to state 3, which is already computed. This illustrates <strong>the</strong> ambiguity in <strong>the</strong> qualitative<br />
calculus. Because <strong>the</strong> length <strong>of</strong> <strong>the</strong> interval above <strong>the</strong> portal on <strong>the</strong> left side is not known, both states 5<br />
and 3 are valid successors <strong>of</strong> state 2. This can be seen in figure 7.4, which depicts a value history<br />
diagram for al five states in this simulation. Ei<strong>the</strong>r amount1 becomes max while ‘amount 2’ reaches<br />
portal place (state 3) or amount 1 becomes max while amount 2 is still above portal place (state 5). In<br />
<strong>the</strong> end, state 5 is also a valid predecessor <strong>of</strong> state 3, which terminates (figure 7.3b).<br />
57
Figure 7.3. state-transition graphs in VisiGarp.<br />
Figure 7.4. Quantity value history.<br />
58
For <strong>the</strong> simulation model <strong>of</strong> vertical differences <strong>of</strong> contained substances, many scenarios’ can be<br />
constructed for all possible configurations <strong>of</strong> containers and levels <strong>of</strong> liquid. Some <strong>of</strong> <strong>the</strong>se scenarios<br />
are suitable to simulating water flows as <strong>the</strong>y occur for example between rivers and seas.<br />
7.2 Simulating <strong>Model</strong> 1<br />
The input scenario for <strong>Model</strong> 1 in figure 7.5 introduces all <strong>the</strong> entities existent in <strong>the</strong> model: <strong>the</strong>re are<br />
two water storages containing water and two air-layers containing vapour. Air-layer 2 also contains <strong>the</strong><br />
entity cloud, <strong>the</strong> equivalent <strong>of</strong> water contained in <strong>the</strong> air. The value for <strong>the</strong> amount <strong>of</strong> water 1 is set to<br />
max and <strong>the</strong> value for amount <strong>of</strong> water 2 is set to plus, all o<strong>the</strong>r substances have <strong>the</strong> value zero for<br />
amount. The quantities pressure for both air-layers are set to plus, thus being equal and in balance.<br />
Both controllers account for <strong>the</strong> continuity <strong>of</strong> values for <strong>the</strong> amount <strong>of</strong> cloud and air-pressure <strong>of</strong> air 2<br />
throughout <strong>the</strong> states where <strong>the</strong>re exist no influences on <strong>the</strong>se quantities.<br />
Figure 7.5. Scenario for <strong>Model</strong> 1.<br />
The simulation <strong>of</strong> <strong>Model</strong> 1 is able to produce a state-transition-graph that activates <strong>the</strong> main processes<br />
<strong>of</strong> <strong>the</strong> water cycle within eight successive states. The model starts with a liquid flow between to water<br />
storages on <strong>the</strong> surface and a heat source heating one <strong>of</strong> <strong>the</strong>m. The simulation continues to find a new<br />
process each successive state and stops when precipitation occurs. Figure 7.6 shows an example<br />
simulation <strong>of</strong> <strong>Model</strong> 1. The states in <strong>the</strong> transition graph are annotated with <strong>the</strong> active process<br />
59
fragment <strong>of</strong> each state. Note that each process remains active throughout <strong>the</strong> simulation once it is<br />
instantiated. The process <strong>of</strong> precipitation in state 61 increases <strong>the</strong> amount <strong>of</strong> water in <strong>the</strong> storage that is<br />
<strong>the</strong> source <strong>of</strong> <strong>the</strong> initial liquid flow in state 1. In between <strong>the</strong>se states, <strong>the</strong> processes evaporation,<br />
transportation and condensation occur. From this perspective, <strong>the</strong> simulation is able to communicate a<br />
causal model that is <strong>of</strong> cyclic nature.<br />
Figure 7.6. Example simulation <strong>of</strong> <strong>Model</strong> 1.<br />
Except for state 1, which has a single successor state, <strong>the</strong> states after state two have a fairly big amount<br />
<strong>of</strong> successors each. Typically in <strong>the</strong> simulation is that each successive state introduces a new process,<br />
which adds more ambiguity by for example instantiating new quantities (e.g. flow rates) with various<br />
possible values. This means that <strong>the</strong> ambiguity created by o<strong>the</strong>r processes that remain active as well is<br />
sometimes multiplied by <strong>the</strong> newly added ambiguity.<br />
The value history diagram <strong>of</strong> <strong>the</strong> simulation (figure 7.7) <strong>of</strong>fers a closer look to <strong>the</strong> values <strong>of</strong> <strong>the</strong><br />
quantities <strong>of</strong> interest in <strong>the</strong> simulation. In state 1 and 2, two processes are active; liquid flow and heat<br />
flow. <strong>Water</strong> flows from water storage 1 to water storage 2, and <strong>the</strong> sun heats <strong>the</strong> latter. In state 4, <strong>the</strong><br />
temperature <strong>of</strong> water 2 reaches max and <strong>the</strong> model fragment ‘evaporation’ increases <strong>the</strong> amount <strong>of</strong><br />
vapour 1 and decreases <strong>the</strong> amount <strong>of</strong> water 2. Because <strong>of</strong> <strong>the</strong> difference in pressure between air 1 and<br />
air 2, <strong>the</strong> fragment ‘transport’ in state 8, exchanges vapour between air 1 and air 2. In state <strong>the</strong> amount<br />
<strong>of</strong> vapour 2 is at plus, and its temperature is below max. Therefore, in state 25, <strong>the</strong> model fragment<br />
‘condensation’ forms a cloud, which stands for water in <strong>the</strong> air. Finally, in state 61, <strong>the</strong> amount <strong>of</strong><br />
clouds becomes plus and <strong>the</strong> fragment precipitation becomes active, which increases <strong>the</strong> amount <strong>of</strong><br />
60
water 1. Theoretically <strong>the</strong> causal chain starts over again, when <strong>the</strong> amount <strong>of</strong> water 1 becomes greater<br />
than <strong>the</strong> amount <strong>of</strong> water 2, similar to state 1.<br />
Figure 7.7. State transition graph in <strong>Model</strong> 1.<br />
Generally speaking, <strong>the</strong> simulation introduces ambiguous behaviour on <strong>the</strong> basis <strong>of</strong> unspecified or<br />
unknown values for quantities. It <strong>the</strong>refore assumes multiple values for <strong>the</strong> quantities. This forms a<br />
drawback in <strong>the</strong> sense that <strong>the</strong> not all simulation results are as intended given <strong>the</strong> model as it is<br />
constructed. The simulation model produces redundant instantiation <strong>of</strong> new quantities <strong>of</strong> <strong>the</strong> same type<br />
and function, each for a different value <strong>of</strong> one or more conditional quantities in a model fragment.<br />
Although this is a well-known effect <strong>of</strong> <strong>the</strong> definition <strong>of</strong> <strong>the</strong> model as it is, <strong>the</strong> ambiguity should be<br />
reduced to that which is implied purely by <strong>the</strong> knowledge in <strong>the</strong> model and leaving out <strong>the</strong> ambiguity<br />
that is generated by undesirable interpretations by <strong>the</strong> algorithm.<br />
However, all resulting states are valid descriptions <strong>of</strong> possible behaviour as can be anticipated from<br />
<strong>the</strong> manner in which <strong>the</strong> system is modelled. The model doesn’t assume values for quantities that are<br />
not appropriate, which would result in erroneous behaviour. For example in figure 7.7, <strong>the</strong> overall<br />
amount <strong>of</strong> substances never increases or decreases, <strong>the</strong>re is always an opposite direction <strong>of</strong> change in<br />
61
one <strong>of</strong> <strong>the</strong> quantities amount (except state 61, termination-state in this simulation). The model can be<br />
qualified as being constructed with <strong>the</strong> purpose to reason about <strong>the</strong> behaviour <strong>of</strong> a water cycle in a<br />
very superficial and abstract way. The model is constructed focusing on <strong>the</strong> occurrence and sequence<br />
<strong>of</strong> certain processes, leaving <strong>the</strong>ir exact functioning implicit, so as <strong>the</strong>ir interdependencies with o<strong>the</strong>r<br />
processes.<br />
7.3 Simulating <strong>Model</strong> 2<br />
Currently <strong>the</strong> simulation model that is used only reasons about liquid flows, heat exchange,<br />
evaporation and vertical movement in <strong>the</strong> atmosphere. The processes condensation and precipitation<br />
have been discussed but are not included in <strong>the</strong> simulation discussed here. The process <strong>of</strong><br />
transportation is not included in <strong>the</strong> model. The simulation model takes as input a scenario <strong>of</strong> <strong>the</strong> water<br />
cycle in which <strong>the</strong> structures sun, seawater, assembly pressure area and assembly atmosphere are<br />
active. The sun and sea engage in a process <strong>of</strong> heat flow because <strong>the</strong> temperature <strong>of</strong> hydrogen is<br />
greater than that <strong>the</strong> temperature <strong>of</strong> <strong>the</strong> sea water. All four o<strong>the</strong>r possible heat flows are assumed to be<br />
non-existent, <strong>the</strong>ir values being equal to each o<strong>the</strong>r. Also <strong>the</strong> movement <strong>of</strong> <strong>the</strong> pressure area is<br />
assumed to be steady, with <strong>the</strong> pressure <strong>of</strong> both assemblies being equal. The scenario is depicted in<br />
figure 7.8.<br />
Figure. 7.8. Scenario for water cycle model 2.<br />
62
In fact <strong>the</strong> simulation starts with three possible initial states from <strong>the</strong> input, because it assumes all<br />
three possibilities with respect to <strong>the</strong> temperature difference between seawater and hydrogen: smaller,<br />
equal and greater. The remaining four heat flow processes between substances are also active, be it<br />
that <strong>the</strong> heat flow is steady (zero). The simulator finds <strong>the</strong> process <strong>of</strong> evaporation in <strong>the</strong> next annotated<br />
state after state “1”. In <strong>the</strong> same state <strong>the</strong> process <strong>of</strong> “pressure area movement” changes from steady to<br />
rising. This is <strong>the</strong> effect <strong>of</strong> <strong>the</strong> inequality between air pressure area and air atmosphere, which is<br />
caused by <strong>the</strong> increase <strong>of</strong> amount <strong>of</strong> vapour. Recall that <strong>the</strong> overall pressure <strong>of</strong> <strong>the</strong> pressure area<br />
assembly is affected by this influence. The causal model associated with this behaviour is shown in<br />
figure 7.10. There are two processes active between <strong>the</strong> pressure area and <strong>the</strong> atmosphere, a energy<br />
exchange as a consequence <strong>of</strong> an inequality in temperature <strong>of</strong> air and a vertical movement on <strong>the</strong> basis<br />
<strong>of</strong> a difference in pressure <strong>of</strong> air. When <strong>the</strong> process <strong>of</strong> evaporation increases <strong>the</strong> amount <strong>of</strong> vapour in<br />
<strong>the</strong> air, <strong>the</strong> systems energy balance start to change: <strong>the</strong> substances contained in <strong>the</strong> pressure area all<br />
exchange heat with each o<strong>the</strong>r and <strong>the</strong>ir net result forms <strong>the</strong> joint influences on <strong>the</strong> quantity pressure <strong>of</strong><br />
<strong>the</strong> pressure area. The successive states in <strong>the</strong> simulation are all characterised by <strong>the</strong> exchange <strong>of</strong><br />
energy between substances and <strong>the</strong> movement <strong>of</strong> <strong>the</strong> pressure area as its consequence. After seven<br />
subsequent states <strong>the</strong> pressure area reaches <strong>the</strong> altitude high. In figure 7.9, a value history diagram is<br />
depicted that shows <strong>the</strong> behaviour <strong>of</strong> several quantities <strong>of</strong> interest. The value history shows that <strong>the</strong><br />
amount <strong>of</strong> water decreases in favour <strong>of</strong> <strong>the</strong> amount <strong>of</strong> vapour. The temperature <strong>of</strong> vapour changes<br />
through continuous energy exchange with o<strong>the</strong>r substances. The movement <strong>of</strong> <strong>the</strong> pressure area<br />
corresponds to flow rate 1, thus <strong>the</strong> processes <strong>of</strong> heat flow and movement are always changing in <strong>the</strong><br />
same direction. The altitude has a single influence from <strong>the</strong> quantity movement and increases in <strong>the</strong><br />
course <strong>of</strong> <strong>the</strong> simulation. The flow rates depicted in <strong>the</strong> figure show <strong>the</strong> effect <strong>of</strong> <strong>the</strong> incoming vapour<br />
and <strong>the</strong> resulting energy exchanges.<br />
63
Figure 7.9. Value history diagram<br />
<strong>Model</strong> 2 <strong>of</strong> <strong>the</strong> water cycle is much more specific in terms <strong>of</strong> dependencies between quantities<br />
throughout <strong>the</strong> whole system. It is also accurate in <strong>the</strong> prediction <strong>of</strong> possible qualitative values and<br />
derivatives for all quantities <strong>of</strong> interest in subsequent states. This is <strong>the</strong> result <strong>of</strong> well-defined<br />
interdependencies between processes and <strong>the</strong>se processes being generic at <strong>the</strong> same time. In <strong>the</strong> model,<br />
quite a few entities in <strong>the</strong> model have shared quantities, so that changes in such shared quantities<br />
propagate throughout big parts <strong>of</strong> <strong>the</strong> system. <strong>An</strong> example <strong>of</strong> this propagation is <strong>the</strong> behaviour<br />
associated with contained gasses in a pressure area. Here, three different gasses share one quantity<br />
pressure. Figure 7.10 shows <strong>the</strong> causal model in state 5 <strong>of</strong> <strong>the</strong> simulation, in which <strong>the</strong> active<br />
influences between <strong>the</strong> entities are depicted.<br />
64
Figure 7.10. Causal model <strong>of</strong> state “5”, generated using VisiGarp.<br />
The generic model fragment “heat flow” is always active between substances that are connected by a<br />
heat path, thus resulting in a complex causal model <strong>of</strong> <strong>the</strong> energy exchange between <strong>the</strong> substances.<br />
When a difference in temperature can be assumed by <strong>the</strong> simulation, ei<strong>the</strong>r <strong>the</strong> subtypes “heat flow<br />
positive” “no flow”or “heat flow negative” become active<br />
When <strong>the</strong> process <strong>of</strong> heat flow occurs between any two substances, <strong>the</strong> amount <strong>of</strong> possible qualitative<br />
states as a result becomes large and increases throughout subsequent states. The point can be<br />
illustrated by examining <strong>the</strong> derivatives <strong>of</strong> <strong>the</strong> quantities ‘flow rate’ that belong to <strong>the</strong> five heat paths<br />
that are defined between <strong>the</strong> substances and shown in figure 7.11 In state 1, <strong>the</strong> flow rate sun to sea<br />
influences <strong>the</strong> temperature <strong>of</strong> <strong>the</strong> water. The same flow rate is also mentioned in <strong>the</strong> model fragment<br />
“Evaporation” where it has an influence on <strong>the</strong> derivative <strong>of</strong> <strong>the</strong> quantity heat. Through <strong>the</strong> defined<br />
dependencies for a closed contained gas, <strong>the</strong> influence propagates to <strong>the</strong> derivative <strong>of</strong> <strong>the</strong> flow rate.<br />
The quantity temperature also propagates this influence to <strong>the</strong> flow rates <strong>of</strong> <strong>the</strong> o<strong>the</strong>r heat paths it is<br />
connected to.<br />
65
Figure 7.11. Value history for flow rates after state “1”.<br />
The qualitative ambiguity produced here is disputable in <strong>the</strong> sense that <strong>the</strong> simulator infers new<br />
behaviour <strong>of</strong> a more global system scope, as a consequence <strong>of</strong> a change that is modelled with local<br />
scope. Apart from <strong>the</strong> validity <strong>of</strong> this behaviour, <strong>the</strong> reason for it to occur in <strong>the</strong> model lies in <strong>the</strong> high<br />
generalisation in <strong>the</strong> models <strong>of</strong> <strong>the</strong> processes. With such a high level <strong>of</strong> generalisation, alterations in<br />
<strong>the</strong> model itself have far-reaching consequences for <strong>the</strong> overall behaviour in <strong>the</strong> model.<br />
The model is thus capable <strong>of</strong> reasoning quite detailed about <strong>the</strong> processes in <strong>the</strong> water cycle that are<br />
highly interrelated per definition. The causal order <strong>of</strong> <strong>the</strong> paths in <strong>the</strong> simulation follow <strong>the</strong> intention<br />
<strong>of</strong> <strong>the</strong> model, thus not introducing illogic behaviour given <strong>the</strong> functioning <strong>of</strong> <strong>the</strong> system. The<br />
simulation however produces a highly ambiguous causal path, for <strong>the</strong> reasons that are discussed<br />
briefly above.<br />
The model is thus currently not suitable for simulating <strong>the</strong> global and interrelated functioning <strong>of</strong> <strong>the</strong><br />
system and keeping <strong>the</strong> amount <strong>of</strong> generated states within reasonable borders at <strong>the</strong> same time. In<br />
order to create manageable simulations, <strong>the</strong> specification <strong>of</strong> multiple modelling assumptions is an<br />
alternative. Such assumption can constrain <strong>the</strong> amount <strong>of</strong> states produced, but can exclude important<br />
behaviour.<br />
<strong>An</strong>alysing <strong>the</strong> result <strong>of</strong> large (qualitative) simulations remains a complex task, which can interfere for<br />
example with <strong>the</strong> effectiveness <strong>of</strong> a particular teaching strategy. <strong>An</strong> improvement <strong>of</strong> this drawback is<br />
that qualitative models can be arranged to structure <strong>the</strong> knowledge into sets <strong>of</strong> smaller simulation<br />
66
models (clusters 3 ), which can support a progressive (learning) route through <strong>the</strong> subjects <strong>of</strong><br />
(educational) interest. Various scenarios that apply to different clusters can be defined which typically<br />
simulate just local parts <strong>of</strong> <strong>the</strong> water cycle. Although this idea <strong>of</strong> curriculum design has been<br />
researched earlier (White & Frederiksen 1990, Winkels & Breuker 1993), an integrated approach on<br />
<strong>the</strong>se ideas specialised for qualitative models is proposed by Salles & Bredeweg (2001), by describing<br />
how to construct progressive learning routes through qualitative simulations in ecology. In <strong>the</strong> paper<br />
article <strong>the</strong>y combine ideas from research mentioned above in order to define a set <strong>of</strong> dimensions on<br />
which partial models can be classified. The classification relies on <strong>the</strong> hierarchical implementation <strong>of</strong><br />
<strong>the</strong> library <strong>of</strong> model fragments, and a set <strong>of</strong> scenario’s that represents a route <strong>of</strong> increasing complexity<br />
through <strong>the</strong> knowledge contained in <strong>the</strong> library.<br />
3 Clusters contain model fragments <strong>of</strong> particular types, e.g. <strong>the</strong> first cluster may contain only single description<br />
views, whereas more complex clusters may contain description views composition views and process views.<br />
67
8. Conclusion and discussion.<br />
The domain <strong>of</strong> <strong>the</strong> water cycle forms a collection <strong>of</strong> physical processes that involve various<br />
components and functioning. Two models <strong>of</strong> <strong>the</strong> water cycle have been developed, each forming a<br />
qualitative ontology <strong>of</strong> <strong>the</strong> domain <strong>of</strong> <strong>the</strong> water cycle. The models vary in <strong>the</strong>ir ontology with respect<br />
to <strong>the</strong> interrelation and generalisation <strong>of</strong> concepts. <strong>Model</strong> 1 gives a high level view on <strong>the</strong> system and<br />
it uses model fragments that are specific for physical situations. <strong>Model</strong> 2 inhibits a more in-depth<br />
specification <strong>of</strong> dependencies that exist within <strong>the</strong> water cycle. In addition, a separate model is<br />
developed for reasoning about vertical differences on <strong>the</strong> surface and corresponding liquid flows. In<br />
terms <strong>of</strong> knowledge articulation, all models add to a multi-level representation <strong>of</strong> <strong>the</strong> water cycle. It is<br />
clear that each model focuses on various aspects and different levels <strong>of</strong> detail, and from <strong>the</strong><br />
perspective <strong>of</strong> multiple scopes <strong>of</strong> human mental models, <strong>the</strong> qualitative models are complementary<br />
The simulations that each model produces also reflect <strong>the</strong> different nature <strong>of</strong> <strong>the</strong> models. <strong>Model</strong> 1<br />
results in a simulation where <strong>the</strong> global behaviour is simulated in a couple <strong>of</strong> successive states, but it<br />
is far from accurate and complete about <strong>the</strong> underlying behaviour, which results in ambiguous<br />
prediction <strong>of</strong> behaviour. The model is constructed with <strong>the</strong> purpose <strong>of</strong> generating a simulation <strong>of</strong> a<br />
particular type, namely a sequence <strong>of</strong> <strong>the</strong> processes <strong>of</strong> <strong>the</strong> water cycle, which also includes <strong>the</strong>ir<br />
general behaviour. The content <strong>of</strong> <strong>the</strong> model is <strong>the</strong>refore adapted to produce <strong>the</strong> type <strong>of</strong> simulation.<br />
<strong>Model</strong> 2 is more generic and complete in its description <strong>of</strong> concepts. It results in a very detailed<br />
simulation, with a high density <strong>of</strong> dependencies. The simulation illustrates <strong>the</strong> approach <strong>of</strong> building a<br />
qualitative model that focuses on a complete representation <strong>of</strong> functioning within <strong>the</strong> domain. As such<br />
<strong>the</strong> simulation has <strong>the</strong> purpose <strong>of</strong> showing how <strong>the</strong> system evolves given its represented functioning.<br />
The result is a causal representation <strong>of</strong> <strong>the</strong> complex interactions <strong>of</strong> <strong>the</strong> water cycle at each point in<br />
time. <strong>Multiple</strong> processes influence a small set <strong>of</strong> entities and <strong>the</strong> simulation predicts valid paths <strong>of</strong><br />
behaviour. But <strong>the</strong> simulation is also highly ambiguous in its prediction <strong>of</strong> behaviour. The reason for<br />
this ambiguity is not inaccuracy or incompleteness, but in this simulation, <strong>the</strong> ambiguity is <strong>the</strong><br />
consequence <strong>of</strong> <strong>the</strong> high density <strong>of</strong> dependencies introduced by <strong>the</strong> relations. On <strong>the</strong> one hand, single<br />
quantities have many active dependencies that result in ambiguous influences. On <strong>the</strong> o<strong>the</strong>r hand<br />
changes in a single quantity sometimes propagate and produce variations in <strong>the</strong> system outside <strong>the</strong><br />
scope <strong>of</strong> <strong>the</strong> process in which <strong>the</strong> quantity participates.<br />
The development <strong>of</strong> large-scale qualitative models confronts <strong>the</strong> developer with trade<strong>of</strong>fs between <strong>the</strong><br />
constraints <strong>of</strong> qualitative simulation algorithms and <strong>the</strong> correct or accurate representation <strong>of</strong> <strong>the</strong><br />
physical situation at hand. Danger exists in making alterations in <strong>the</strong> domain knowledge contained by<br />
68
<strong>the</strong> model for <strong>the</strong> sake <strong>of</strong> constructing efficient and more insightful simulations. Large qualitative<br />
models are sometimes not effective in <strong>the</strong>ir capability <strong>of</strong> showing system behaviour by simulation, by<br />
creating complex and ambiguous causal paths. At least for large and generic models like <strong>Model</strong> 2,<br />
involving various different interrelated processes, <strong>the</strong> simulation models can be divided into pieces<br />
that simulate just parts <strong>of</strong> <strong>the</strong> system. Decomposition on <strong>the</strong> basis <strong>of</strong> a curriculum for didactic use and<br />
<strong>the</strong> use <strong>of</strong> modelling assumptions also make <strong>the</strong> simulation <strong>of</strong> such models more insightful to <strong>the</strong><br />
observer.<br />
Throughout this research, issues have been raised about <strong>the</strong> relation between qualitative models and<br />
<strong>the</strong>ir purpose in educational contexts. Whereas research has shown that <strong>the</strong>re exist reasons to assume<br />
that <strong>the</strong> use <strong>of</strong> qualitative models and simulations has important connections to human reasoning, <strong>the</strong>re<br />
is no sound evidence that <strong>the</strong> approach facilitates knowledge acquisition per se. From a didactic<br />
perspective, <strong>the</strong> research on teaching and human learning <strong>of</strong>fers knowledge about styles and methods<br />
<strong>of</strong> knowledge communication and knowledge acquisition. Much <strong>of</strong> this knowledge is still to be<br />
researched with respect to implementation in automated teaching devices. The models in this research<br />
are aimed at communicating conceptual and causal knowledge <strong>of</strong> an ecological domain. They form a<br />
basis from which a curriculum may be produced that is able to guide learners through <strong>the</strong> domain <strong>of</strong><br />
<strong>the</strong> water cycle while successively addressing deeper levels and higher complexity. Given an explicit<br />
definition <strong>of</strong> <strong>the</strong> didactic purpose that should be covered by <strong>the</strong> qualitative models, an appropriate<br />
means <strong>of</strong> interacting with <strong>the</strong> knowledge contained by <strong>the</strong> models needs to be developed. Such an<br />
interaction can for example take <strong>the</strong> form <strong>of</strong> constructivism, where subjects engage in a process <strong>of</strong><br />
knowledge articulation and learning by doing.<br />
The qualitative models here form a multi-level knowledge representation that can be expanded with<br />
additional A.I. techniques in order to engage in teaching activities. Examples <strong>of</strong> <strong>the</strong>se techniques are<br />
(automatic) question generation and explanation generation. <strong>An</strong>o<strong>the</strong>r role <strong>of</strong> <strong>the</strong> discussed models in<br />
this context is that <strong>the</strong>y can serve as norm models in support <strong>of</strong> learners model building activities.<br />
Feedback and questions should be generated on <strong>the</strong> basis <strong>of</strong> comparison <strong>of</strong> models to <strong>the</strong> norm model.<br />
The future scope <strong>of</strong> <strong>the</strong> current work lies in <strong>the</strong> integration <strong>of</strong> qualitative water cycle models from<br />
different perspectives into simulation models, using multiple scenarios for each perspective on <strong>the</strong><br />
system. Specific knowledge in terms <strong>of</strong> curriculum design for <strong>the</strong> domain <strong>of</strong> <strong>the</strong> water cycle can help<br />
classifying <strong>the</strong> knowledge into educationally relevant knowledge chunks or dimensions. The resulting<br />
classification can be mapped to scenarios with explicit didactic purposes. The possibility <strong>of</strong> integrating<br />
multiple levels <strong>of</strong> analysis and exploring different causal paths <strong>of</strong> physical systems is what makes<br />
qualitative models promising with respect to knowledge communication. The possibility <strong>of</strong> using <strong>the</strong>se<br />
models in an automated environment that is able to support didactic methods by including specific<br />
69
types <strong>of</strong> interaction (e.g. curriculum planning, automated feedback, explanation) makes <strong>the</strong> use <strong>of</strong><br />
qualitative models promising with respect to instruction.<br />
In order to assess <strong>the</strong> facilitating effect on learning <strong>of</strong> interaction with qualitative models, various<br />
types <strong>of</strong> evaluation can be conducted. These evaluations are usually aimed at investigating <strong>the</strong> type <strong>of</strong><br />
interaction with qualitative models and relate this to subsequent increase <strong>of</strong> knowledge <strong>of</strong> learners.<br />
Such an evaluation would be appropriate when additional aspects are included in <strong>the</strong> qualitative<br />
models and simulations, such as subject matter sequencing and providing feedback and explanation to<br />
<strong>the</strong> learners. Purely concentrating on qualitative model building and simulations, performed by ei<strong>the</strong>r<br />
domain experts or learners, a logic next step is to integrate model building environments with (visual)<br />
simulation (inspection) tools. The testing <strong>of</strong> models and <strong>the</strong>refore also <strong>the</strong> development cycle <strong>of</strong><br />
qualitative models for simulation purposes can be supported to a better extent. A benchmark in which<br />
both models can be defined and simulated, for example should allow to document, compare, and edit<br />
different versions <strong>of</strong> models and <strong>the</strong>ir simulation histories in an integrated manner.<br />
9. Acknowledgements<br />
This study would not exist without <strong>the</strong> intensive co-operation with Dr. Bert Bredeweg at <strong>the</strong><br />
department <strong>of</strong> Social Science Informatics <strong>of</strong> <strong>the</strong> University <strong>of</strong> Amsterdam. His ideas and contribution<br />
in modelling activities have been <strong>of</strong> great support to this work. Fur<strong>the</strong>rmore <strong>the</strong> exchange <strong>of</strong> ideas and<br />
co-operation with Barbara Dubbeldam, Maarten van Ho<strong>of</strong>, Roland Groen, and <strong>An</strong>ders Bouwer were <strong>of</strong><br />
great value.<br />
70
Literature<br />
Aleven, V., & Koedinger, K.R. (2000). The Need For Tutorial Dialog To Support Self-Explanation.<br />
In C.P. Rose & R. Freedman (Eds.), Building Dialogue Systems for Tutorial Applications, Papers <strong>of</strong><br />
<strong>the</strong> 2000 AAAI Fall Symposium (pp. 65-73). Technical Report FS-00-01. Menlo Park, CA: AAAI<br />
Press.<br />
<strong>An</strong>derson, J. R. (1993). Rules <strong>of</strong> <strong>the</strong> Mind. Hillsdale, NJ: Erlbaum<br />
<strong>An</strong>derson, J., Boyle, C., Farrell, R. & Reiser, B. (1987). Cognitive principles in <strong>the</strong> design <strong>of</strong> computer<br />
tutors. In P. Morris (ed.), <strong>Model</strong>ing Cognition. NY: John Wiley.<br />
Bredeweg. B. (1992). Expertise in Qualitative Prediction <strong>of</strong> Behaviour. Ph.D. <strong>the</strong>sis, University <strong>of</strong><br />
Amsterdam, Amsterdam, The Ne<strong>the</strong>rlands.<br />
Bredeweg B. and Schut C.. (1993). Building Qualitative <strong>Model</strong>s for Reasoning about Device<br />
Behavior. Social Science Computer Review, pages 350-365, volume 11, number 3, Duke University<br />
Press, Durham, USA.<br />
Bredeweg B., Winkels, R. (1998). Qualitative <strong>Model</strong>s in Interactive Learning Environments: an<br />
Introduction. Interactive Learning Environments, pages 1-18, Volume 5, (introduction to special<br />
issue).<br />
Breuker, J. (1994b). A suite <strong>of</strong> problem types. In Breuker, J. & van de Velde, W., editors,<br />
CommonKADSLibrary for Expertise <strong>Model</strong>ing, pages 57--87. Amsterdam, The Ne<strong>the</strong>rlands, IOS<br />
Press.<br />
Brown, J. S., Burton R. R., & Johan deKleer (1982). Pedagogical, natural language and knowledge<br />
engineering techniques in SOPHIE I, II and III. In D. Sleeman & J. S. Brown (Eds.), Intelligent<br />
Tutoring Systems (pp. 227-282). New York: Academic Press.<br />
Brown, D.C. (1996) CS 540: Artificial Intelligence in Design, web page,<br />
http://cs.wpi.edu/Research/aidg/CS540/aid.html, Computer Science Dept., WPI, 1996.<br />
Collins, A., & Gentner, D. (1983). <strong>Multiple</strong> models <strong>of</strong> evaporation processes. Proceedings <strong>of</strong> <strong>the</strong> Fifth<br />
<strong>An</strong>nual Conference <strong>of</strong> <strong>the</strong> Cognitive Science Society.<br />
71
Collins, J. and K. Forbus. Building qualitative models <strong>of</strong> <strong>the</strong>rmodynamic processes. Proceedings <strong>of</strong> <strong>the</strong><br />
Qualitative Reasoning Workshop. 1989.<br />
Carbonell, J. R. (1970b). Mixed-initiative man-computer instructional dialogues (BBN Report No.<br />
1971). Cambridge, MA: Bolt Beranek and Newman Inc.<br />
Goel, A. K. (1992) Representation <strong>of</strong> Design Functions in Experienced-Based Design. In: Intelligent<br />
Computer Aided Design, (Eds) D. C. Brown, M. Waldron & H. Yoshikawa, Elsevier Science<br />
Publishers (North-Holland), , pp. 283-303<br />
Grecu, D. L. & Brown, D. C. (1996)"Learning by Single Function Agents During Spring Design",<br />
Artificial Intelligence in Design `96, (Eds.) J. S. Gero & F. Sudweeks, Kluwer Academic Publishers,<br />
June 1996, pp. 409-428<br />
John W. Collins, Kenneth D. Forbus (1987): Reasoning about Fluids via Molecular Collections.<br />
AAAI 1987: 590-594<br />
Collins A and Stevens A., (1977). The goal structure <strong>of</strong> a socratic tutor. Association for Computing<br />
Machinery <strong>An</strong>nual Conference, 1977. (Also available as BBN Report No. 3518 from Bolt Beranek<br />
and Newman Inc., Cambridge, Mass., 02138).<br />
De Kleer, J. and Brown, J. S. (1984). A Qualitative physics based on confluences, Artificial<br />
Intelligence, 24, p.7-83.<br />
Falkenhainer, B., Forbus, K. D., & Gentner, D. (1986). The structure-mapping engine. Proceedings <strong>of</strong><br />
<strong>the</strong> American Association for Artificial Intelligence (pp. 272--277), Philadelphia.<br />
Falkenhainer B., Forbus, K.D (1991). Compositional <strong>Model</strong>ing: Finding <strong>the</strong> Right <strong>Model</strong> for <strong>the</strong> Job.<br />
Artificial Intelligence 51(1-3): 95-143 (1991)<br />
Forbus, K. D. (1984). Qualitative Process Theory. Artificial Intelligence, 1984. In: Bobrow, 1984.<br />
Forbus, K. D. and Gentner, D. (1986). Causal reasoning about quantities. Proceedings Cognitive<br />
Science Society, pages 196 -- 207.<br />
72
Forbus, K. D. and Gentner, D. (1997). Qualitative mental <strong>Model</strong>s: Simulations or Memories?<br />
Proceedings <strong>of</strong> <strong>the</strong> Eleventh International Workshop on Qualitative Reasoning, Cortona, Italy.<br />
Forbus, KD., (1988). Qualitative physics: Past, present, and future. in Exploring Artificial<br />
Intelligence, Morgan-Kaufmann.<br />
Forbus, K.D., (2001). <strong>Articulate</strong> s<strong>of</strong>tware for science and engineering education. In Forbus, K.,<br />
Feltovich, P., and Canas, A. (Eds) Smart machines in education: The coming revolution in educational<br />
technology. AAAI Press.<br />
Forbus, K.D., and Whalley, P.B., (1994). Using qualitative physics to build articulate s<strong>of</strong>tware for<br />
<strong>the</strong>rmodynamics education. Proceedings <strong>of</strong> <strong>the</strong> 12th National Conference on Artificial Intelligence.<br />
Forbus K.D., Gentner, D. Arthur B. Markman, Ronald W. Ferguson: <strong>An</strong>alogy just looks like high level<br />
perception: why a domain-general approach to analogical mapping is right. 231-257<br />
D. Grecu, D.C. Brown "Dimensions <strong>of</strong> Learning in Design ", Artificial Intelligence for Engineering<br />
Design, <strong>An</strong>alysis and Manufacturing, special issue on "Machine Learning in Design", A.H.B. Duffy,<br />
D.C. Brown & A.K. Goel (eds.), Cambridge University Press, April 1998, 12, pp. 117-122<br />
Guerrin, F., (1990). Qualitative reasoning about an ecological process: interpretation in<br />
hydroecology. Ecological <strong>Model</strong>ling 59: 165-201<br />
Hayes, P. J., (1978). The naive physics manifesto. In D. Michie, editor, Expert Systems in <strong>the</strong> Micro-<br />
Electronic Age. Edinburgh University Press<br />
Hayes,P.J., (1978), Naive physics I: Ontology for liquids. University <strong>of</strong> Geneva, ISSCo, Working<br />
Paper 35, Geneva, Switzerland,.<br />
Hayes, P.J., (1985), Naive Physics I: Ontology for Liquids, Formal Theories <strong>of</strong> <strong>the</strong> Commonsense<br />
World, J. Hobbs and R. Moore, eds., Ablex, Norwood, NJ, 1985, pp. 71-108.<br />
Heller, U., Struss, P., Guerrin, F., (1995), and W. Roque. A qualitative modeling approach to algal<br />
bloom prediction. Working Papers <strong>of</strong> <strong>the</strong> IJCAI-Workshop on Artificial Intelligence and <strong>the</strong><br />
Environment, Montreal, Canada, 1995.<br />
Iwasaki, Y., Simon., H.A., (1986), Causality in device behavior. Artificial Intelligence, 29(1):3—32.<br />
73
Kim, H. (1993): Qualitative Reasoning about <strong>the</strong> geometry <strong>of</strong> fluid flow. Proceedings <strong>of</strong> <strong>the</strong> Twelfth<br />
<strong>An</strong>nual Conference <strong>of</strong> <strong>the</strong> Cognitive Science Society, Cambridge, MA, July, 1990<br />
Koedinger, K.R., Daniel D. Su<strong>the</strong>rs, Kenneth D. Forbus: Component-Based Construction <strong>of</strong> a Science<br />
Learning Space. Intelligent Tutoring Systems 1998: 166-175<br />
Kuipers, Benjamin: Qualitative Simulation. Artificial Intelligence 29(3): 289-338 (1986)<br />
Laurillard, Diana. Rethinking University Teaching: A Conversational Framework for <strong>the</strong> Effective Use<br />
<strong>of</strong> Learning Technologies', 2nd Ed, 2002<br />
Lee C-L., Iyengar G. and Kota, S. (1992) Automated Configuration Design <strong>of</strong> Hydraulic Systems. In:<br />
Artificial Intelligence in Design'92, (Ed) J. S. Gero, Kluwer Academic Publishers, , pp. 61-82.<br />
Lesgold, A. (1988). Toward a <strong>the</strong>ory <strong>of</strong> curriculum for use in designing intelligent instructional<br />
systems. In H. Mandl & A. Lesgold (Eds.), Learning issues for intelligent tutoring systems. New York:<br />
Springer-Verlag.<br />
McCarthy, J. (1963). A Basis for a Ma<strong>the</strong>matical Theory <strong>of</strong> Computation. In Braffort, P. and<br />
Hirschberg, D., editors, Computer Programming and Formal Systems, pages 33-70. North-Holland,<br />
Amsterdam.<br />
McCarthy, John and P.J. Hayes (1969): ``Some Philosophical Problems from <strong>the</strong> Standpoint <strong>of</strong><br />
Artificial Intelligence'', in D. Michie (ed), Machine Intelligence 4, American Elsevier, New York,<br />
NY,. Reprinted in [McCarthy, 1990].<br />
Mizoguchi, R. & Bourdeau J. (2000) "Using Ontological Engineering to Overcome Common<br />
AI-ED Problems." Journal <strong>of</strong> Artificial Intelligence and Education<br />
Nayak, P. P. Context-Dependent Behaviors: A Preliminary Report. In: Intelligent Computer Aided<br />
Design, (Eds) D. C. Brown, M. Waldron & H. Yoshikawa, Elsevier Science Publishers (North-<br />
Holland), 1992, pp. 237-251<br />
Newell, A. (1982). The Knowledge Level. Artificial Intelligence, 1(18):87--127.<br />
Rittle-Johnson, B., Koedinger, K.R. (2001) Proceedings <strong>of</strong> <strong>the</strong> Cognitive Using cognitive models to<br />
guide instructional design: The case <strong>of</strong> fraction division. Science Society Conference<br />
74
Ritter, S. and Koedinger, K. R.: <strong>An</strong> architecture for plug-in tutor agents. Journal <strong>of</strong> Artificial<br />
Intelligence in Education 7, 3/4 (1996) 315-347<br />
Quillian, M. R. (1968). Semantic memory. In M. Minsky (Ed.), Semantic infomation<br />
Paulo Salles, Helen Pain and Robert I. Muetzelfeldt Qualitative ecological models for tutoring<br />
systems: a comparative study, University <strong>of</strong> Edinburgh, UK<br />
Paulo Salles and Bert Bredeweg. 2001. Constructing Progressive Learning Routes through Qualitative<br />
Simulation <strong>Model</strong>s in Ecology. Proceedings <strong>of</strong> <strong>the</strong> International workshop on Qualitative Reasoning,<br />
QR'01, pages 82-89, San <strong>An</strong>tonio, Texas, USA, May 17-19, Gautam Biswas (editor).<br />
Jacobijn Sandberg and Jerry <strong>An</strong>driessen (1999). Where is education heading and what about A.I.?<br />
International Journal <strong>of</strong> Artificial Intelligence in Education, 1999, 10, 135-150<br />
C. Schut and B. Bredeweg. (1996) <strong>An</strong> Overview <strong>of</strong> Approaches to Qualitative <strong>Model</strong> Construction.<br />
The Knowledge Engineering Review, volume 11, number 1, pages 1-25<br />
Sowa, J.F. (2000). Knowledge representation Logical, Philosophical and Computional Foudations.<br />
Brooks-Cole<br />
Stefik, M (1981) Planning with Constraints (MOLGEN: Part 1). Artificial Intelligence, Vol. 16, No.<br />
2, North-Holland, 1981, pp. 111-140<br />
Tong, C. (1987) AI In Engineering Design. Artificial Intelligence in Engineering 2,3<br />
Y. Umeda & T. Tomiyama (1997), Functional Reasoning in Design, IEEE Expert, Vol. 12, No.2,<br />
March-April 1997, pp. 42-48.<br />
Wenger, E. (1987). Artificial Intelligence and Tutoring Systems. San Mateo, CA: Morgan Kaufmann.<br />
White, B., & Frederiksen, J. (1986). Intelligent tutoring systems based on qualitative model evolutions.<br />
In Proceedings <strong>of</strong> <strong>the</strong> National Conference on Artificial Intelligence.<br />
75
White, B., & Frederiksen, J. "Causal <strong>Model</strong> Progressions as a Foundation for Intelligent Learning<br />
Environments." Artificial Intelligence, 24, 99-157, 1990.<br />
F. Zhao, ``Intelligent Simulation in Designing Complex Dynamical Control Systems'', Book chapter in<br />
Artificial Intelligence in Industrial Decision Making, Control, and Automation, Tzafestas and<br />
Verbruggen (eds.), pp. 127-158, Kluwer, 1995<br />
76